This year’s findings reveal something we have all been sensing: the threat of landscape is not just evolving—it is accelerating. AI has fundamentally changed the equation, impacting the speed, scale, and sophistication of cyberattacks in ways that render many traditional defensive assumptions obsolete. Yet AI also represents our most powerful tool for adaptation.

Understanding the acceleration

The metrics tell a stark story, but the operational implications matter more. We’re observing cyberattacks that execute in the time it takes a user to click—ClickFix techniques that bypass layered defenses through social engineering at machine speed. In cloud environments, the window between deployment and compromise has collapsed to 48 hours for containers, fundamentally challenging our assumptions about hardening timelines.

The economics have shifted as well. AI-powered phishing campaigns now achieve 50 times profitability improvements by automating personalization at scale. We’re tracking North Korean operations that have embedded tens of thousands of workers globally, turning the remote workforce into a persistent cyberthreat vector. This is not opportunistic. Indeed, it is industrial-scale infiltration.

The sophistication curve continues its steep climb. Our telemetry shows an 87% increase in disruptive campaigns targeting Microsoft Azure environments. Credential theft attempts are up 23%, data exfiltration up 58%. We are now tracking early indicators of autonomous malware capable of lateral movement and adaptive behavior without human direction.

What strikes me most is the operational coordination. Through Microsoft Threat Intelligence, we are observing campaigns spanning more than 130 countries where nation-states, criminal syndicates, and commercial mercenaries share infrastructure and tactics. Access brokers have created marketplaces that blur lines between espionage and crime. The models–scalable, resilient, and disturbingly efficient.

From threat awareness to strategic action

Here is the paradox every CISO faces: threats are accelerating, yet our defensive capabilities have never been stronger. The gap is not technology. The gap is in how we think about and operationalize security. Legacy approaches that separate security from business strategy, that prioritize prevention over resilience, that treat threat incidents as failures rather than inevitable events—these mindsets are now liabilities.

The path forward requires fundamental shifts:

Security as a business enabler, not a control point. We just embed security into every business process, from product development to supply chain management. When security becomes integral to how organizations operate, rather than a gate they must pass through, we move faster while managing risk more effectively. This is not about lowering standards. This is about building security into the foundation rather than adding it as a façade.

Resilience as the primary objective. The question isn’t if an incident will occur, but how quickly we can detect, contain, and recover from it. When cyberattacks execute in seconds and compromises happen within 48 hours, our response capabilities must match that velocity. This means tested playbooks, empowered teams, and automated response mechanisms that operate at machine speed.

Intelligence and automation as force multipliers. The same AI technologies that let cyberattackers scale operations can amplify our defense capabilities—if we deploy them strategically. Automation is not about replacing security teams. It is about letting them operate at the speed and scale that modern threats demand.

The evolved CISO mandate

The role of the CISO has fundamentally expanded. We are no longer purely technologists. We are risk managers, strategic advisors, and organizational change agents. The board needs us to translate technical cyberthreats into business risks and resilience strategies into competitive advantages.

This evolution demands new capabilities:

Cross-functional leadership that transcends IT. When a social engineering attack can compromise an organization in seconds, response requires coordinated actions across IT, legal, human resources, communications, and executive leadership. We must build these partnerships before the crisis, not during it.

Continuous adaptation as operational discipline. The 48-hour container compromise window and the instant infection vectors we are seeing mean that continuous monitoring, regular testing, and rapid iteration are not best practices. They are survival requirements. Our defenses, policies, and response capabilities must evolve as quicky as threats.

Governance that anticipates regulatory evolution. As governments increase transparency requirements and impose consequences for malicious activity, we must ensure our organizations can meet both the letter and the spirit of emerging regulations. This includes understanding third-party risks, from access brokers to embedded cyberthreats in our workforce and supply chains.

Proven strategies for operationalizing security resilience

From our work with customers, own operational experience, and implementation of the Secure Future Initiative (SFI), three priorities rise to the top:

Modern identity controls are non-negotiable. With 97% of identity attacks targeting passwords, phishing-resistant MFA fundamentally alters the risk equation. This isn’t about adding layers—it’s about eliminating entire attack vectors. Organizations that deploy phishing-resistant authentication see dramatic reductions in successful compromises.

Incident response readiness determines outcome. When attacks move at machine speed, response time becomes the critical variable. This means regular simulations, tested playbooks, and teams empowered to act decisively. We must practice for the scenarios we’ll face, not the ones we hope to avoid. The organizations that recover fastest are those that have failed in simulation and learned before the real event.

Collective defense is no longer optional. Against campaigns spanning more than 130 countries and cyberattacker ecosystems sharing infrastructure, isolated defense is ineffective. Intelligence sharing, collaborative best practices, and sector-wide coordination are force multipliers that benefit everyone. The cyberthreats we face are too sophisticated and too coordinated for any organization to defend alone.

We’ve been applying these same principles internally through our Secure Future Initiative. Rather than keep our implementation lessons internal, we’re publishing the actual patterns and practices we’ve used—the specific approaches that worked, the trade-offs we encountered, and the practical steps other organizations can adapt. The SFI patterns and practices library includes detailed guidance on challenges like securing multi-tenant environments, protecting software supply chains, and implementing Zero Trust for source code access.

What I appreciate about these patterns is that they are written by practitioners who have actually implemented them. Each one outlines the problem, explains how we solved it internally at Microsoft, and provides recommendations that you can evaluate for your own environment. No glossy overviews—just the operating details of what worked and what did not.

Steps to strengthen resilience and response across your organization 

The acceleration we are witnessing—cyberattack speed, operational scale, and technical sophistication—demands an equivalent acceleration in our response. This is not about working harder; it’s about working differently. It means treating AI and automation as operational imperatives, not future projects. It means building identity security as foundational infrastructure, not a compliance checkbox. It means developing incident response capabilities that match the velocity of modern cyberattacks.

Most fundamentally, it means embracing our evolved role as CISOs. We are architects of organizational resilience in an era where cyberthreats move at machine speed and span continents. This requires equal parts of technical depth, strategic vision, and collaborative leadership.

The cyberthreat landscape will continue to evolve. Our mandate is to evolve faster, to build organizations that are not just secure but resilient, adaptive, and prepared for whatever comes next. That is the challenge facing every CISO today. It is also the opportunity to build something stronger than what came before.

For a detailed and comprehensive analysis, explore the full Microsoft Digital Defense Report 2025.

Microsoft Deputy CISOs

To hear more from Microsoft Deputy CISOs, check out the OCISO blog series.

To stay on top of important security industry updates, explore resources specifically designed for CISOs, and learn best practices for improving your organization’s security posture, join the Microsoft CISO Digest distribution list.

Learn more with Microsoft Security

To learn more about Microsoft Security solutions, visit our website. Bookmark the Security blog to keep up with our expert coverage on security matters. Also, follow us on LinkedIn (Microsoft Security) and X (@MSFTSecurity) for the latest news and updates on cybersecurity.

The post The CISO imperative: Building resilience in an era of accelerated cyberthreats appeared first on Microsoft Security Blog.

Microsoft maintains a continuous effort to protect its platforms and customers from fraud and abuse. From blocking imposters on Microsoft Azure and adding anti-scam features to Microsoft Edge, to fighting tech support fraud with new features in Windows Quick Assist, this edition of Cyber Signals takes you inside the work underway and important milestones achieved that protect customers.

We are all defenders. 

Between April 2024 and April 2025, Microsoft:

• Thwarted $4 billion in fraud attempts.
• Rejected 49,000 fraudulent partnership enrollments.
• Blocked about 1.6 million bot signup attempts per hour.

The evolution of AI-enhanced cyber scams

AI has started to lower the technical bar for fraud and cybercrime actors looking for their own productivity tools, making it easier and cheaper to generate believable content for cyberattacks at an increasingly rapid rate. AI software used in fraud attempts runs the gamut, from legitimate apps misused for malicious purposes to more fraud-oriented tools used by bad actors in the cybercrime underground.

AI tools can scan and scrape the web for company information, helping cyberattackers build detailed profiles of employees or other targets to create highly convincing social engineering lures. In some cases, bad actors are luring victims into increasingly complex fraud schemes using fake AI-enhanced product reviews and AI-generated storefronts, where scammers create entire websites and e-commerce brands, complete with fake business histories and customer testimonials. By using deepfakes, voice cloning, phishing emails, and authentic-looking fake websites, threat actors seek to appear legitimate at wider scale.

According to the Microsoft Anti-Fraud Team, AI-powered fraud attacks are happening globally, with much of the activity coming from China and Europe, specifically Germany due in part to Germany’s status as one of the largest e-commerce and online services markets in the European Union (EU). The larger a digital marketplace in any region, the more likely a proportional degree of attempted fraud will take place.

E-commerce fraud

Fraudulent e-commerce websites can be set up in minutes using AI and other tools requiring minimal technical knowledge. Previously, it would take threat actors days or weeks to stand up convincing websites. These fraudulent websites often mimic legitimate sites, making it challenging for consumers to identify them as fake. 

Using AI-generated product descriptions, images, and customer reviews, customers are duped into believing they are interacting with a genuine merchant, exploiting consumer trust in familiar brands.

AI-powered customer service chatbots add another layer of deception by convincingly interacting with customers. These bots can delay chargebacks by stalling customers with scripted excuses and manipulating complaints with AI-generated responses that make scam sites appear professional.

In a multipronged approach, Microsoft has implemented robust defenses across our products and services to protect customers from AI-powered fraud. Microsoft Defender for Cloud provides comprehensive threat protection for Azure resources, including vulnerability assessments and threat detection for virtual machines, container images, and endpoints.

Microsoft Edge features website typo protection and domain impersonation protection using deep learning technology to help users avoid fraudulent websites. Edge has also implemented a machine learning-based Scareware Blocker to identify and block potential scam pages and deceptive pop-up screens with alarming warnings claiming a computer has been compromised. These attacks try to frighten users into calling fraudulent support numbers or downloading harmful software.

Job and employment fraud

The rapid advancement of generative AI has made it easier for scammers to create fake listings on various job platforms. They generate fake profiles with stolen credentials, fake job postings with auto-generated descriptions, and AI-powered email campaigns to phish job seekers. AI-powered interviews and automated emails enhance the credibility of job scams, making it harder for job seekers to identify fraudulent offers.

To prevent this, job platforms should introduce multifactor authentication for employer accounts to make it harder for bad actors to take over legitimate hirers’ listings and use available fraud-detection technologies to catch suspicious content.

Fraudsters often ask for personal information, such as resumes or even bank account details, under the guise of verifying the applicant’s information. Unsolicited text and email messages offering employment opportunities that promise high pay for minimal qualifications are typically an indicator of fraud.

Employment offers that include requests for payment, offers that seem too good to be true, unsolicited offers or interview requests over text message, and a lack of formal communication platforms can all be indicators of fraud.

Tech support scams

Tech support scams are a type of fraud where scammers trick victims into unnecessary technical support services to fix a device or software problems that don’t exist. The scammers may then gain remote access to a computer—which lets them access all information stored on it, and on any network connected to it or install malware that gives them access to the computer and sensitive data.

Tech support scams are a case where elevated fraud risks exist, even if AI does not play a role. For example, in mid-April 2024, Microsoft Threat Intelligence observed the financially motivated and ransomware-focused cybercriminal group Storm-1811 abusing Windows Quick Assist software by posing as IT support. Microsoft did not observe AI used in these attacks; Storm-1811 instead impersonated legitimate organizations through voice phishing (vishing) as a form of social engineering, convincing victims to grant them device access through Quick Assist. 

Quick Assist is a tool that enables users to share their Windows or macOS device with another person over a remote connection. Tech support scammers often pretend to be legitimate IT support from well-known companies and use social engineering tactics to gain the trust of their targets. They then attempt to employ tools like Quick Assist to connect to the target’s device. 

Quick Assist and Microsoft are not compromised in these cyberattack scenarios; however, the abuse of legitimate software presents risk Microsoft is focused on mitigating. Informed by Microsoft’s understanding of evolving cyberattack techniques, the company’s anti-fraud and product teams work closely together to improve transparency for users and enhance fraud detection techniques. 

The Storm-1811 cyberattacks highlight the capability of social engineering to circumvent security defenses. Social engineering involves collecting relevant information about targeted victims and arranging it into credible lures delivered through phone, email, text, or other mediums. Various AI tools can quickly find, organize, and generate information, thus acting as productivity tools for cyberattackers. Although AI is a new development, enduring measures to counter social engineering attacks remain highly effective. These include increasing employee awareness of legitimate helpdesk contact and support procedures, and applying Zero Trust principles to enforce least privilege across employee accounts and devices, thereby limiting the impact of any compromised assets while they are being addressed. 

Microsoft has taken action to mitigate attacks by Storm-1811 and other groups by suspending identified accounts and tenants associated with inauthentic behavior. If you receive an unsolicited tech support offer, it is likely a scam. Always reach out to trusted sources for tech support. If scammers claim to be from Microsoft, we encourage you to report it directly to us at http://approjects.co.za/?big=reportascam. 

Building on the Secure Future Initiative (SFI), Microsoft is taking a proactive approach to ensuring our products and services are “Fraud-resistant by Design.” In January 2025, a new fraud prevention policy was introduced: Microsoft product teams must now perform fraud prevention assessments and implement fraud controls as part of their design process. 

Recommendations

• Strengthen employer authentication: Fraudsters often hijack legitimate company profiles or create fake recruiters to deceive job seekers. To prevent this, job platforms should introduce multifactor authentication and Verified ID as part of Microsoft Entra ID for employer accounts, making it harder for unauthorized users to gain control.
• Monitor for AI-based recruitment scams: Companies should deploy deepfake detection algorithms to identify AI-generated interviews where facial expressions and speech patterns may not align naturally.
• Be cautious of websites and job listings that seem too good to be true: Verify the legitimacy of websites by checking for secure connections (https) and using tools like Microsoft Edge’s typo protection.
• Avoid providing personal information or payment details to unverified sources: Look for red flags in job listings, such as requests for payment or communication through informal platforms like text messages, WhatsApp, nonbusiness Gmail accounts, or requests to contact someone on a personal device for more information.

Using Microsoft’s security signal to combat fraud

Microsoft is actively working to stop fraud attempts using AI and other technologies by evolving large-scale detection models based on AI, such as machine learning, to play defense by learning from and mitigating fraud attempts. Machine learning is the process that helps a computer learn without direct instruction using algorithms to discover patterns in large datasets. Those patterns are then used to create a comprehensive AI model, allowing for predictions with high accuracy.

We have developed in-product safety controls that warn users about potential malicious activity and integrate rapid detection and prevention of new types of attacks.

Our fraud team has developed domain impersonation protection using deep-learning technology at the domain creation stage, to help protect against fraudulent e-commerce websites and fake job listings. Microsoft Edge has incorporated website typo protection, and we have developed AI-powered fake job detection systems for LinkedIn.

Microsoft Defender Smartscreen is a cloud-based security feature that aims to prevent unsafe browsing habits by analyzing websites, files, and applications based on their reputation and behavior. It is integrated into Windows and the Edge browser to help protect users from phishing attacks, malicious websites, and potentially harmful downloads.

Furthermore, Microsoft’s Digital Crimes Unit (DCU) partners with others in the private and public sector to disrupt the malicious infrastructure used by criminals perpetuating cyber-enabled fraud. The team’s longstanding collaboration with law enforcement around the world to respond to tech support fraud has resulted in hundreds of arrests and increasingly severe prison sentences worldwide. The DCU is applying key learnings from past actions to disrupt those who seek to abuse generative AI technology for malicious or fraudulent purposes. 

Quick Assist features and remote help combat tech support fraud

To help combat tech support fraud, we have incorporated warning messages to alert users about possible tech support scams in Quick Assist before they grant access to someone approaching them purporting to be an authorized IT department or other support resource.

Windows users must read and click the box to acknowledge the security risk of granting remote access to the device.

Microsoft has significantly enhanced Quick Assist protection for Windows users by leveraging its security signal. In response to tech support scams and other threats, Microsoft now blocks an average of 4,415 suspicious Quick Assist connection attempts daily, accounting for approximately 5.46% of global connection attempts. These blocks target connections exhibiting suspicious attributes, such as associations with malicious actors or unverified connections.

Microsoft’s continual focus on advancing Quick Assist safeguards seeks to counter adaptive cybercriminals, who previously targeted individuals opportunistically with fraudulent connection attempts, but more recently have sought to target enterprises with more organized cybercrime campaigns that Microsoft’s actions have helped disrupt.

Our Digital Fingerprinting capability, which leverages AI and machine learning, drives these safeguards by providing fraud and risk signals to detect fraudulent activity. If our risk signals detect a possible scam, the Quick Assist session is automatically ended. Digital Fingerprinting works by collecting various signals to detect and prevent fraud.

For enterprises combating tech support fraud, Remote Help is another valuable resource for employees. Remote Help is designed for internal use within an organization and includes features that make it ideal for enterprises.

By reducing scams and fraud, Microsoft aims to enhance the overall security of its products and protect its users from malicious activities.

Consumer protection tips

Fraudsters exploit psychological triggers such as urgency, scarcity, and trust in social proof. Consumers should be cautious of:

• Impulse buying—Scammers create a sense of urgency with “limited-time” deals and countdown timers.
• Trusting fake social proof—AI generates fake reviews, influencer endorsements, and testimonials to appear legitimate.
• Clicking on ads without verification—Many scam sites spread through AI-optimized social media ads. Consumers should cross-check domain names and reviews before purchasing.
• Ignoring payment security—Avoid direct bank transfers or cryptocurrency payments, which lack fraud protections.

Job seekers should verify employer legitimacy, be on the lookout for common job scam red flags, and avoid sharing personal or financial information with unverified employers.

• Verify employer legitimacy—Cross-check company details on LinkedIn, Glassdoor, and official websites to verify legitimacy.
• Notice common job scam red flags—If a job requires upfront payments for training materials, certifications, or background checks, it is likely a scam. Unrealistic salaries or no-experience-required remote positions should be approached with skepticism. Emails from free domains (such as johndoehr@gmail.com instead of hr@company.com) are also typically indicators of fraudulent activity.
• Be cautious of AI-generated interviews and communications—If a video interview seems unnatural, with lip-syncing delays, robotic speech, or odd facial expressions, it could be deepfake technology at work. Job seekers should always verify recruiter credentials through the company’s official website before engaging in any further discussions.
• Avoid sharing personal or financial information—Under no circumstances should you provide a Social Security number, banking details, or passwords to an unverified employer.

Microsoft is also a member of the Global Anti-Scam Alliance (GASA), which aims to bring governments, law enforcement, consumer protection organizations, financial authorities and providers, brand protection agencies, social media, internet service providers, and cybersecurity companies together to share knowledge and protect consumers from getting scammed.

Recommendations

• Remote Help: Microsoft recommends using Remote Help instead of Quick Assist for internal tech support. Remote Help is designed for internal use within an organization and incorporates several features designed to enhance security and minimize the risk of tech support hacks. It is engineered to be used only within an organization’s tenant, providing a safer alternative to Quick Assist.
• Digital Fingerprinting: This identifies malicious behaviors and ties them back to specific individuals. This helps in monitoring and preventing unauthorized access.
• Blocking full control requests: Quick Assist now includes warnings and requires users to check a box acknowledging the security implications of sharing their screen. This adds a layer of helpful “security friction” by prompting users who may be multitasking or preoccupied to pause to complete an authorization step.

Kelly Bissell: A cybersecurity pioneer combating fraud in the new era of AI

Kelly Bissell’s journey into cybersecurity began unexpectedly in 1990. Initially working in computer science, Kelly was involved in building software for healthcare patient accounting and operating systems at Medaphis and Bellsouth, now AT&T.

His interest in cybersecurity was sparked when he noticed someone logged into a phone switch attempting to get free long-distance calls and traced the intruder back to Romania. This incident marked the beginning of Kelly’s career in cybersecurity.

“I stayed in cybersecurity hunting for bad actors, integrating security controls for hundreds of companies, and helping shape the NIST security frameworks and regulations such as FFIEC, PCI, NERC-CIP,” he explains.

Currently, Kelly is Corporate Vice President of Anti-Fraud and Product Abuse within Microsoft Security. Microsoft’s fraud team employs machine learning and AI to build better detection code and understand fraud operations. They use AI-powered solutions to detect and prevent cyberthreats, leveraging advanced fraud detection frameworks that continuously learn and evolve.

“Cybercrime is a trillion-dollar problem, and it’s been going up every year for the past 30 years. I think we have an opportunity today to adopt AI faster so we can detect and close the gap of exposure quickly. Now we have AI that can make a difference at scale and help us build security and fraud protections into our products much faster.”

Previously Kelly managed the Microsoft Detection and Response Team (DART) and created the Global Hunting, Oversight, and Strategic Triage (GHOST) team that detected and responded to attackers such as Storm-0558 and Midnight Blizzard.

Prior to Microsoft, during his time at Accenture and Deloitte, Kelly collaborated with companies and worked extensively with government agencies like the Department of Homeland Security’s Cybersecurity and Infrastructure Security Agency (CISA) and the Federal Bureau of Investigation, where he helped build security systems inside their operations.

His time as Chief Information Security Officer (CISO) at a bank exposed him to addressing both cybersecurity and fraud, leading to his involvement in shaping regulatory guidelines to protect banks and eventually Microsoft.

Kelly has also played a significant role in shaping regulations around the National Institute of Standards and Technology (NIST) and Payment Card Industry (PCI) compliance, which helps ensure the security of businesses’ credit card transactions, among others.

Internationally, Kelly played a crucial role in helping establish agencies and improve cybersecurity measures. As a consultant in London, he helped stand up the United Kingdom’s National Cyber Security Centre (NCSC), which is part of the Government Communications Headquarters (GCHQ), the equivalent of CISA. Kelly’s efforts in content moderation with several social media companies, including YouTube, were instrumental in removing harmful content.

That’s why he’s excited about Microsoft’s partnership with GASA. GASA brings together governments, law enforcement, consumer protection organizations, financial authorities, internet service providers, cybersecurity companies, and others to share knowledge and define joint actions to protect consumers from getting scammed.

“If I protect Microsoft, that’s good, but it’s not sufficient. In the same way, if Apple does their thing, and Google does their thing, but if we’re not working together, we’ve all missed the bigger opportunity. We must share cybercrime information with each other and educate the public. If we can have a three-pronged approach of tech companies building security and fraud protection into their products, public awareness, and sharing cybercrime and fraudster information with law enforcement, I think we can make a big difference,” he says.

Next steps with Microsoft Security

To learn more about Microsoft Security solutions, visit our website. Bookmark the Security blog to keep up with our expert coverage on security matters. Also, follow us on LinkedIn (Microsoft Security) and X (@MSFTSecurity) for the latest news and updates on cybersecurity.

Methodology: Microsoft platforms and services, including Azure, Microsoft Defender for Office, Microsoft Threat Intelligence, and Microsoft Digital Crimes Unit (DCU), provided anonymized data on threat actor activity and trends. Additionally, Microsoft Entra ID provided anonymized data on threat activity, such as malicious email accounts, phishing emails, and attacker movement within networks. Additional insights are from the daily security signals gained across Microsoft, including the cloud, endpoints, the intelligent edge, and telemetry from Microsoft platforms and services. The $4 billion figure represents an aggregated total of fraud and scam attempts against Microsoft and our customers in consumer and enterprise segments (in 12 months).

The post Cyber Signals Issue 9 | AI-powered deception: Emerging fraud threats and countermeasures appeared first on Microsoft Security Blog.

The evolution of ransomware attacks

Modern ransomware campaigns are meticulously planned. Cyberattackers understand that their chances of securing a ransom increase significantly if they can inflict widespread damage across a victim’s environment. The rationale is simple: paying the ransom becomes the most viable option when the alternative—restoring the environment and recovering data—is technically unfeasible, time-consuming, and costly.

This level of damage happens in minutes and even seconds, where bad actors embed themselves within an organization’s environment, laying the groundwork for a coordinated cyberattack that can encrypt dozens, hundreds, or even thousands of devices within minutes. To execute such a campaign, threat actors must overcome several challenges such as evading protection, mapping the network, maintaining their code execution ability, and preserving persistency in the environment, building their way to securing two major prerequisites necessary to execute ransomware on multiple devices simultaneously:

• High-privilege accounts: Whether cyberattackers choose to drop files and encrypt the devices locally or perform remote operations over the network, they must obtain the ability to authenticate to a device. In an on-premises environment, cyberattackers usually target domain admin accounts or other high-privilege accounts, as those can authenticate to the most critical resources in the environment.
• Access to central network assets: To execute the ransomware attack as fast and as wide as possible, threat actors aim to achieve access to a central asset in the network that is exposed to many endpoints. Thus, they can leverage the possession of high-privilege accounts and connect to all devices visible in their line of sight.

The role of domain controllers in ransomware campaigns

Domain controllers are the backbone of any on-premises environment, managing identity and access through Active Directory (AD). They play a pivotal role in enabling cyberattackers to achieve their goals by fulfilling two critical requirements:

1. Compromising highly privileged accounts

Domain controllers house the AD database, which contains sensitive information about all user accounts, including highly privileged accounts like domain admins. By compromising a domain controller, threat actors can:

• Extract password hashes: Dumping the NTDS.dit file allows cyberattackers to obtain password hashes for every user account.
• Create and elevate privileged accounts: Cyberattackers can generate new accounts or manipulate existing ones, assigning them elevated permissions, ensuring continued control over the environment.

With these capabilities, cyberattackers can authenticate as highly privileged users, facilitating lateral movement across the network. This level of access enables them to deploy ransomware on a scale, maximizing the impact of their attack.

2. Exploiting centralized network access

Domain controllers handle crucial tasks like authenticating users and devices, managing user accounts and policies, and keeping the AD database consistent across the network. Because of these important roles, many devices need to interact with domain controllers regularly to ensure security, efficient resource management, and operational continuity. That’s why domain controllers need to be central in the network and accessible to many endpoints, making them a prime target for cyberattackers looking to cause maximum damage with ransomware attacks.

Given these factors, it’s no surprise that domain controllers are frequently at the center of ransomware operations. Cyberattackers consistently target them to gain privileged access, move laterally, and rapidly deploy ransomware across an environment. We’ve seen in more than 78% of human-operated cyberattacks, threat actors successfully breach a domain controller. Additionally, in more than 35% of cases, the primary spreader device—the system responsible for distributing ransomware at scale—is a domain controller, highlighting its crucial role in enabling widespread encryption and operational disruption.

Case study: Ransomware attack using a compromised domain controller

In one notable case, a small-medium manufacturer fell victim to a well-known, highly skilled threat actor attempting to execute a widespread Akira ransomware attack:

Pre domain-compromise activity

After gaining initial access, presumably through leveraging the customer’s VPN infrastructure, and prior to obtaining domain admin privileges, the cyberattackers initiated a series of actions focused on mapping potential assets and escalating privileges. A wide, remote execution of secrets dump is detected on Microsoft Defender for Endpoint-onboarded devices and User 1 (domain user) is contained by attack disruption.

Post domain-compromise activity

Once securing domain admin (User 2) credentials, potentially through leveraging the victim’s non-onboarded estate, the attacker immediately attempts to connect to the victim’s domain controller (DC1) using Remote Desktop Protocol (RDP) from the cyberattacker’s controlled device. When gaining access to DC1, the cyberattacker leverages the device to perform the following set of actions:

• Reconnaissance—The cyberattacker leverages the domain controller’s wide network visibility and high privileges to map the network using different tools, focusing on servers and network shares.
• Defense evasion—Leveraging the domain controller’s native group policy functionality, the cyberattacker attempts to tamper with the victim’s antivirus by modifying security-related group policy settings.
• Persistence—The cyberattacker leverages the direct access to Active Directory, creating new domain users (User 3 and User 4) and adding them to the domain admin group, thus establishing a set of highly privileged users that would later on be used to execute the ransomware attack.

Encryption over the network

Once the cyberattacker takes control over a set of highly privileged users, this provides them access to any domain-joined resource, including comprehensive network access and visibility. It will also allow them to set up tools for the encryption phase of the cyberattack.

Assuming they’re able to validate a domain controller’s effectiveness, they begin by running the payload locally on the domain controller. Attack disruption detects the threat actor’s attempt to run the payload and contains User 2, User 3, and the cyberattacker-controlled device used to RDP to the domain controller.

After successfully containing Users 2 and 3, the cyberattacker proceeded to log in to the domain controller using User 4, who had not yet been utilized. After logging into the device, the cyberattacker attempted to encrypt numerous devices over the network from the domain controller, leveraging the access provided by User 4.

Attack disruption detects the initiation of encryption over the network and automatically granularly contains device DC1 and User 4, blocking the attempted remote encryption on all Microsoft Defender for Endpoint-onboarded and targeted devices.

Protecting your domain controllers

Given the central role of domain controllers in ransomware attacks, protecting them is critical to preventing large-scale damage. However, securing domain controllers is particularly challenging due to their fundamental role in network operations. Unlike other endpoints, domain controllers must remain highly accessible to authenticate users, enforce policies, and manage resources across the environment. This level of accessibility makes it difficult to apply traditional security measures without disrupting business continuity. Hence, security teams constantly face the complex challenge of striking the right balance between security and operational functionality.

To address this challenge, Defender for Endpoint introduced contain high value assets (HVA), an expansion of our contain device capability designed to automatically contain HVAs like domain controllers in a granular manner. This feature builds on Defender for Endpoint’s capability to classify device roles and criticality levels to deliver a custom, role-based containment policy, meaning that if a sensitive device, such a domain controller, is compromised, it is immediately contained in less than three minutes, preventing the cyberattacker from moving laterally and deploying ransomware, while at the same time maintaining the operational functionality of the device. The ability of the domain controller to distinguish between malicious and benign behavior helps keep essential authentication and directory services up and running. This approach provides rapid, automated cyberattack containment without sacrificing business continuity, allowing organizations to stay resilient against sophisticated human-operated cyberthreats.

Now your organization’s domain controllers can leverage automatic attack disruption as an extra line of defense against malicious actors trying to overtake high value assets and exert costly ransomware attacks.

Learn more

Explore these resources to stay updated on the latest automatic attack disruption capabilities:

• Learn more about Microsoft Defender for Endpoint.
• Read about the new containment features.
• Learn how attack disruption safeguards your domain controllers in this video.
• Check out our latest infographic.
• Read about automatic attack disruption.

To learn more about Microsoft Security solutions, visit our website. Bookmark the Security blog to keep up with our expert coverage on security matters. Also, follow us on LinkedIn (Microsoft Security) and X (@MSFTSecurity) for the latest news and updates on cybersecurity.

¹Average cost per data breach in the United States 2006-2024, Ani Petrosyan. October 10, 2024.

The post How cyberattackers exploit domain controllers using ransomware appeared first on Microsoft Security Blog.

Prior to 2020, there was an unspoken rule of threat actors to not launch attacks against schools and children, infrastructure, and healthcare organizations.¹ However, that “rule” no longer applies, and in the past four years the healthcare threat landscape has seen tremendous shifts for the worse.

To put this shift into context, consider these trends from the Microsoft Threat Intelligence report showing healthcare cybersecurity challenges:

• Healthcare is one of the top 10 most targeted industries in the second quarter of 2024²—and has been for the past four quarters.
• Ransomware attacks are costly, with healthcare organizations losing an average of $900,000 per day on downtime alone.³
• In a recent study, out of the 99 healthcare organizations that admitted to paying a ransom and disclosed the ransom paid, the average payment was $4.4 million.⁴

The serious impact of ransomware on healthcare

While the potential financial risk for healthcare organizations is high, lives are at stake because ransomware attacks impact patient outcomes. If healthcare providers are not able to use diagnostic equipment or access patient medical records because it’s under ransom, care will be disrupted.

Healthcare facilities located near hospitals that are impacted by ransomware are also affected because they experience a surge of patients needing care and are unable to support them in an urgent manner. As a result, patients can experience longer wait times, which studies show could lead to more severe stroke cases and heart attack cases.⁵

These attacks don’t just impact facilities in large cities; in fact, rural health clinics are also a target for cyberattacks. They are particularly vulnerable to ransomware incidents because they often have limited means to prevent and remediate security risks. This can be devastating for a community as these hospitals are often the only healthcare option for many miles in the communities they serve.  

Why healthcare is an appealing target for threat actors

Healthcare organizations collect and store extremely sensitive data, which likely contributes to threat actors targeting them in ransomware attacks. However, a more significant reason these facilities are at risk is the potential for huge financial payouts. As referenced earlier, lives are at stake and healthcare facilities committed to patient care can’t risk poor patient outcomes if their systems are taken down. They also can’t risk their patients’ data being exposed if they don’t pay the ransom. That reputation for paying ransoms—for understandable reasons—makes them a target.

Healthcare facilities are also targeted because of their limited security resources and cybersecurity investments to defend against these threats compared to other sectors. Facilities often lack staff dedicated to cybersecurity and in fact, some facilities don’t have a chief information security officer (CISO) or dedicated security operations center at all. Instead, their IT department may be tasked with managing cybersecurity. Doctors, nurses, and healthcare staff may not have received any cybersecurity training or know the signs to look for to identify a phishing email.

How cyber criminals target healthcare organizations

Financially motivated cyber criminals are using an evolving set of ransomware tactics on healthcare organizations. One common approach involves two steps. First, they gain access to an organization’s network, often using social engineering tactics through a phishing email or text. Then, they use that access to deploy ransomware to encrypt and lock healthcare systems and data so they can seek a ransom for their release.

“Once ransomware is deployed, attackers typically move quickly to encrypt critical systems and data, often within a matter of hours,” said Jack Mott of Microsoft Threat Intelligence in the Microsoft ransomware report. “They target essential infrastructure, such as patient records, diagnostic systems, and even billing operations, to maximize the impact and pressure on healthcare organizations to pay the ransom.”

Social engineering tactics often involve convincing the email recipient to act in ways they normally wouldn’t, such as clicking on an unknown link, and using the tactics of urgency, emotion, and habit. Social engineering fraud is a serious problem. In just this fiscal year, a staggering 389 healthcare institutions across the United States fell victim to ransomware attacks, according to the 2024 Microsoft Digital Defense Report.⁶ The aftermath was severe, resulting in network closures, offline systems, delays in critical medical operations, and rescheduled appointments.

Another common approach is ransomware as a service (RaaS), a cybercrime business model growing in popularity. The RaaS model is an agreement between an operator, who develops extortion tools, and an affiliate, who deploys the ransomware. Both parties benefit from a successful ransomware and extortion attack, and it’s “democratized access to sophisticated ransomware tools,” Mott said. This model enables cyber criminals without the means of developing their own tools to launch their nefarious activities. Sometimes, they may simply purchase network access from a cybercrime group that has already breached a network. RaaS severely widens the risk to healthcare organizations, making ransomware more accessible and frequent.

Cybercrime tactics continue to grow in sophistication. Microsoft is continually tracking the latest cybercrime threats to support our customers and increase the knowledge of the entire global community. These threats include actions by threat actor groups Vanilla Tempest and Sangria Tempest, which are known for their financially motivated criminal activities.

Take a collective defense approach to boost your cyber resilience and visibility

We recognize that not all organizations have a robust cybersecurity team or even the resources to enable a cybersecurity resilience strategy. This is why it is important for us as a community to come together and share best practices, tools, and guidance. We encourage your organization to collaborate with regional, national, and global healthcare organizations such as Health-ISAC (Information Sharing and Analysis Centers). The Health-ISAC provides healthcare organizations with platforms to exchange threat intelligence. Health-ISAC Chief Security Officer Errol Weiss says these organizations are like “virtual neighborhood watch programs,” sharing threat experiences and defense strategies. 

It’s also important to foster a security-first mindset among healthcare staff. Dr. Christian Dameff and Dr. Jeff Tully, Co-directors of the University of California San Diego Center for Healthcare Cybersecurity, emphasize that breaking down silos between IT security teams, emergency managers, and clinical staff to develop cohesive incident response plans is key. They also recommend running high-fidelity clinical simulations that expose doctors and nurses to real-world cyberattack scenarios.

For rural hospitals that provide critical services to the communities they serve across the US, Microsoft created the Microsoft Cybersecurity Program for Rural Hospitals, which provides affordable access to Microsoft security solutions, builds cybersecurity capacity, and helps solve root challenges through innovation.

For healthcare organizations that have the resources, as part of this report we provide guidance on how to:

• Establish a robust governance framework.
• Create an incident response and detection plan. Then be prepared to execute it efficiently during an actual attack to minimize damage and ensure a quick recovery.
• Implement continuous monitoring and real-time detection capabilities.
• Educate your organization using our cybersecurity awareness and education #BeCyberSmart Kit.
• Harness more resilience strategies found in the report.

Given the serious cyberthreats against healthcare organizations, it’s critical to protect your assets by understanding the situation and taking steps to prevent it. For more details on the current healthcare cyberthreat landscape and ransomware threats, and for more in-depth guidance on boosting resilience, read the “US healthcare at risk: Strengthening resiliency against ransomware attacks” report and watch our healthcare threat intelligence briefing video, which is included in the report. To stay up-to-date on the latest threat intelligence insights and get actionable guidance for your security efforts, bookmark Microsoft Security Insider.

Learn more

To learn more about Microsoft Security solutions, visit our website. Bookmark the Security blog to keep up with our expert coverage on security matters. Also, follow us on LinkedIn (Microsoft Security) and X (@MSFTSecurity) for the latest news and updates on cybersecurity.

¹How to protect your networks from ransomware, justice.gov.

²Threat Landscape: Healthcare and Public Health Sector, April 2024. Microsoft Threat Intelligence.

³On average, healthcare organizations lose $900,000 per day to downtime from ransomware attacks, Comparitech. March 6, 2024.

⁴Healthcare Ransomware Attacks Continue to Increase in Number and Severity, The HIPAA Journal. September 2024.

⁵Ransomware Attack Associated With Disruptions at Adjacent Emergency Departments in the US, JAMA Network. May 8, 2023.

⁶Microsoft Digital Defense Report 2024.

The post Microsoft Threat Intelligence healthcare ransomware report highlights need for collective industry action appeared first on Microsoft Security Blog.

Education is essentially an “industry of industries,” with K-12 and higher education enterprises handling data that could include health records, financial data, and other regulated information. At the same time, their facilities can host payment processing systems, networks that are used as internet service providers (ISPs), and other diverse infrastructure. The cyberthreats that Microsoft observes across different industries tend to be compounded in education, and threat actors have realized that this sector is inherently vulnerable. With an average of 2,507 cyberattack attempts per week, universities are prime targets for malware, phishing, and IoT vulnerabilities.¹ 

Security staffing and IT asset ownership also affect education organizations’ cyber risks. School and university systems, like many enterprises, often face a shortage of IT resources and operate a mix of both modern and legacy IT systems. Microsoft observes that in the United States, students and faculty are more likely to use personal devices in education compared to Europe, for example. Regardless of ownership however, in these and other regions, busy users do not always have a security mindset. 

This edition of Cyber Signals delves into the cybersecurity challenges facing classrooms and campuses, highlighting the critical need for robust defenses and proactive measures. From personal devices to virtual classes and research stored in the cloud, the digital footprint of school districts, colleges, and universities has multiplied exponentially.  

We are all defenders. 

A uniquely valuable and vulnerable environment 

The education sector’s user base is very different from a typical large commercial enterprise. In the K-12 environment, users include students as young as six years old. Just like any public or private sector organization, there is a wide swath of employees in school districts and at universities including administration, athletics, health services, janitorial, food service professionals, and others. Multiple activities, announcements, information resources, open email systems, and students create a highly fluid environment for cyberthreats.

Virtual and remote learning have also extended education applications into households and offices. Personal and multiuser devices are ubiquitous and often unmanaged—and students are not always cognizant about cybersecurity or what they allow their devices to access.

Education is also on the front lines confronting how adversaries test their tools and their techniques. According to data from Microsoft Threat Intelligence, the education sector is the third-most targeted industry, with the United States seeing the greatest cyberthreat activity.

Cyberthreats to education are not only a concern in the United States. According to the United Kingdom’s Department of Science Innovation and Technology 2024 Cybersecurity Breaches Survey, 43% of higher education institutions in the UK reported experiencing a breach or cyberattack at least weekly.² 

QR codes provide an easily disguised surface for phishing cyberattacks

Today, quick response (QR) codes are quite popular—leading to increased risks of phishing cyberattacks designed to gain access to systems and data. Images in emails, flyers offering information about campus and school events, parking passes, financial aid forms, and other official communications all frequently contain QR codes. Physical and virtual education spaces might be the most “flyer friendly” and QR code-intensive environments anywhere, given how big a role handouts, physical and digital bulletin boards, and other casual correspondence help students navigate a mix of curriculum, institutional, and social correspondence. This creates an attractive backdrop for malicious actors to target users who are trying to save time with a quick image scan. 

Recently the United States Federal Trade Commission issued a consumer alert on the rising threat of malicious QR codes being used to steal login credentials or deliver malware.³

Microsoft Defender for Office 365 telemetry shows that approximately more than 15,000 messages with malicious QR codes are targeted toward the educational sector daily—including phishing, spam, and malware. 

Legitimate software tools can be used to quickly generate QR codes with embedded links to be sent in email or posted physically as part of a cyberattack. And those images are hard for traditional email security solutions to scan, making it even more important for faculty and students to use devices and browsers with modern web defenses. 

Targeted users in the education sector may use personal devices without endpoint security. QR codes essentially enable the threat actor to pivot to these devices. QR code phishing (since its purpose is to target mobile devices) is compelling evidence of mobile devices being used as an attack vector into enterprises—such as personal accounts and bank accounts—and the need for mobile device protection and visibility. Microsoft has significantly disrupted QR code phishing attacks. This shift in tactics is evident in the substantial decrease in daily phishing emails intercepted by our system, dropping from 3 million in December 2023 to just 179,000 by March 2024. 

Universities present their own unique challenges. Much of university culture is based on collaboration and sharing to drive research and innovation. Professors, researchers, and other faculty operate under the notion that technology, science—simply knowledge itself—should be shared widely. If someone appearing as a student, peer, or similar party reaches out, they’re often willing to discuss potentially sensitive topics without scrutinizing the source. 

University operations also span multiple industries. University presidents are effectively CEOs of healthcare organizations, housing providers, and large financial organizations—the industry of industries factor, again. Therefore, top leaders can can be prime targets for anyone attacking those sectors.

The combination of value and vulnerability found in education systems has attracted the attention of a spectrum of cyberattackers—from malware criminals employing new techniques to nation-state threat actors engaging in old-school spy craft.  

Microsoft continually monitors threat actors and threat vectors worldwide. Here are some key issues we’re seeing for education systems. 

Email systems in schools offer wide spaces for compromise 

The naturally open environment at most universities forces them to be more relaxed in their email hygiene. They have a lot of emails amounting to noise in the system, but are often operationally limited in where and how they can place controls, because of how open they need to be for alumni, donors, external user collaboration, and many other use cases.  

Education institutions tend to share a lot of announcements in email. They share informational diagrams around local events and school resources. They commonly allow external mailers from mass mailing systems to share into their environments. This combination of openness and lack of controls creates a fertile ground for cyberattacks.

AI is increasing the premium on visibility and control  

Cyberattackers recognizing higher education’s focus on building and sharing can survey all visible access points, seeking entry into AI-enabled systems or privileged information on how these systems operate. If on-premises and cloud-based foundations of AI systems and data are not secured with proper identity and access controls, AI systems become vulnerable. Just as education institutions adapted to cloud services, mobile devices and hybrid learning—which introduced new waves of identities and privileges to govern, devices to manage, and networks to segment—they must also adapt to the cyber risks of AI by scaling these timeless visibility and control imperatives.

Nation-state actors are after valuable IP and high-level connections 

Universities handling federally funded research, or working closely with defense, technology, and other industry partners in the private sector, have long recognized the risk of espionage. Decades ago, universities focused on telltale physical signs of spying. They knew to look for people showing up on campus taking pictures or trying to get access to laboratories. Those are still risks, but today the dynamics of digital identity and social engineering have greatly expanded the spy craft toolkit. 

Universities are often epicenters of highly sensitive intellectual property. They may be conducting breakthrough research. They may be working on high-value projects in aerospace, engineering, nuclear science, or other sensitive topics in partnership with multiple government agencies.  

For cyberattackers, it can be easier to first compromise somebody in the education sector who has ties to the defense sector and then use that access to more convincingly phish a higher value target.  

Universities also have experts in foreign policy, science, technology, and other valuable disciplines, who may willingly offer intelligence, if deceived in social-engineering cyberattacks employing false or stolen identities of peers and others who appear to be in individuals’ networks or among trusted contacts. Apart from holding valuable intelligence themselves, compromised accounts of university employees can become springboards into further campaigns against wider government and industry targets.

Nation-state actors targeting education 

Peach Sandstorm

Peach Sandstorm has used password spray attacks against the education sector to gain access to infrastructure used in those industries, and Microsoft has also observed the organization using social engineering against targets in higher education.  

Mint Sandstorm 

Microsoft has observed a subset of this Iranian attack group targeting high-profile experts working on Middle Eastern affairs at universities and research organizations. These sophisticated phishing attacks used social engineering to compel targets to download malicious files including a new, custom backdoor called MediaPl. 

Mabna Institute  

In 2023, the Iranian Mabna Institute conducted intrusions into the computing systems of at least 144 United States universities and 176 universities in 21 other countries.  

The stolen login credentials were used for the benefit of Iran’s Islamic Revolutionary Guard Corps and were also sold within Iran through the web. Stolen credentials belonging to university professors were used to directly access university library systems. 

Emerald Sleet

This North Korean group primarily targets experts in East Asian policy or North and South Korean relations. In some cases, the same academics have been targeted by Emerald Sleet for nearly a decade.  

Emerald Sleet uses AI to write malicious scripts and content for social engineering, but these attacks aren’t always about delivering malware. There’s also an evolving trend where they simply ask experts for policy insight that could be used to manipulate negotiations, trade agreements, or sanctions. 

Moonstone Sleet 

Moonstone Sleet is another North Korean actor that has been taking novel approaches like creating fake companies to forge business relationships with educational institutions or a particular faculty member or student.  

One of the most prominent attacks from Moonstone Sleet involved creating a fake tank-themed game used to target individuals at educational institutions, with a goal to deploy malware and exfiltrate data. 

Storm-1877  

This actor largely engages in cryptocurrency theft using a custom malware family that they deploy through various means. The ultimate goal of this malware is to steal crypto wallet addresses and login credentials for crypto platforms.  

Students are often the target for these attacks, which largely start on social media. Storm-1877 targets students because they may not be as aware of digital threats as professionals in industry. 

A new security curriculum 

Due to education budget and talent constraints and the inherent openness of its environment, solving education security is more than a technology problem. Security posture management and prioritizing security measures can be a costly and challenging endeavor for these institutions—but there is a lot that school systems can do to protect themselves.  

Maintaining and scaling core cyberhygiene will be key to securing school systems. Building awareness of security risks and good practices at all levels—students, faculty, administrators, IT staff, campus staff, and more—can help create a safer environment.  

For IT and security professionals in the education sector, doing the basics and hardening the overall security posture is a good first step. From there, centralizing the technology stack can help facilitate better monitoring of logging and activity to gain a clearer picture into the overall security posture and any vulnerabilities. 

Oregon State University 

Oregon State University (OSU), an R1 research-focused university, places a high priority on safeguarding its research to maintain its reputation. In 2021, it experienced an extensive cybersecurity incident unlike anything before. The cyberattack revealed gaps in OSU’s security operations.

“The types of threats that we’re seeing, the types of events that are occurring in higher education, are much more aggressive by cyber adversaries.”

—David McMorries, Chief Information Security Officer at Oregon State University

In response to this incident, OSU created its Security Operations Center (SOC), which has become the centerpiece of the university’s security effort. AI has also helped automate capabilities and helped its analysts, who are college students, learn how to quickly write code—such as threat hunting with more advanced hunting queries. 

Arizona Department of Education 

A focus on Zero Trust and closed systems is an area that the Arizona Department of Education (ADE) takes further than the state requirements. It blocks all traffic from outside the United States from its Microsoft 365 environment, Azure, and its local datacenter.

“I don’t allow anything exposed to the internet on my lower dev environments, and even with the production environments, we take extra care to make sure that we use a network security group to protect the app services.”

—Chris Henry, Infrastructure Manager at the Arizona Department of Education 

Follow these recommendations:  

• The best defense against QR code attacks is to be aware and pay attention. Pause, inspect the code’s URL before opening it, and don’t open QR codes from unexpected sources, especially if the message uses urgent language or contains errors. 

• Consider implementing “protective domain name service,” a free tool that helps prevent ransomware and other cyberattacks by blocking computer systems from connecting to harmful websites. Prevent password spray attacks with a stringent password and deploy multifactor authentication.  

• Educate students and staff about their security hygiene, and encourage them to use multifactor authentication or passwordless protections. Studies have shown that an account is more than 99.9% less likely to be compromised when using multifactor authentication.   

• Microsoft has launched role-based trainings for leaders, educators, students, parents, and IT professionals aligned to recommendations of the United States Cybersecurity and Infrastructure Security Agency.    

Corey Lee has always had an interest in solving puzzles and crimes. He started his college career at Penn State University in criminal justice, but soon realized his passion for digital forensics after taking a course about investigating a desktop computer break-in.  

After completing his degree in security and risk analysis, Corey came to Microsoft focused on gaining cross-industry experience. He’s worked on securing everything from federal, state, and local agencies to commercial enterprises, but today he focuses on the education sector.  

After spending time working across industries, Corey sees education through a different lens—the significantly unique industry of industries. The dynamics at play inside the education sector include academic institutions, financial services, critical infrastructure like hospitals and transportation, and partnerships with government agencies. According to Corey, working in such a broad field allows him to leverage skillsets from multiple industries to address specific problems across the landscape. 

The fact that education could also be called underserved from a cybersecurity standpoint is another compelling challenge, and part of Corey’s personal mission. The education industry needs cybersecurity experts to elevate the priority of protecting school systems. Corey works across the public and industry dialogue, skilling and readiness programs, incident response, and overall defense to protect not just the infrastructure of education, but students, parents, teachers, and staff. 

Today, Corey is focused reimagining student security operations centers, including how to inject AI into the equation and bring modern technology and training to the table. By growing the cybersecurity work force in education and giving them new tools, he’s working to elevate security in the sector in a way that’s commensurate with how critical the industry is for the future. 

Next steps with Microsoft Security

To learn more about Microsoft Security solutions, visit our website. Bookmark the Security blog to keep up with our expert coverage on security matters. Also, follow us on LinkedIn (Microsoft Security) and X (@MSFTSecurity) for the latest news and updates on cybersecurity.

¹Global Cyberattacks Continue to Rise with Africa and APAC Suffering Most, Check Point Blog. April 27, 2023.

²Cyber security breaches survey 2024: education institutions annex, The United Kingdom Department for Science, Innovation & Technology. April 9, 2024

³Scammers hide harmful links in QR codes to steal your information, Federal Trade Commission (Alvaro Puig), December 6, 2023.

Methodology: Snapshot and cover stat data represent telemetry from Microsoft Defender for Office 365 showing how a QR code phishing attack was disrupted by image detection technology and how Security Operations teams can respond to this threat. Platforms like Microsoft Entra provided anonymized data on threat activity, such as malicious email accounts, phishing emails, and attacker movement within networks. Additional insights are from the 78 trillion daily security signals processed by Microsoft each day, including the cloud, endpoints, the intelligent edge, and telemetry from Microsoft platforms and services including Microsoft Defender. Microsoft categorizes threat actors into five key groups: influence operations; groups in development; and nation-state, financially motivated, and private sector offensive actors. The new threat actors naming taxonomy aligns with the theme of weather.  

© 2024 Microsoft Corporation. All rights reserved. Cyber Signals is for informational purposes only. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS DOCUMENT. This document is provided “as is.” Information and views expressed in this document, including URL and other Internet website references, may change without notice. You bear the risk of using it. This document does not provide you with any legal rights to any intellectual property in any Microsoft product. 

The post ​​Cyber Signals Issue 8 | Education under siege: How cybercriminals target our schools​​ appeared first on Microsoft Security Blog.

We also made a commitment to identify and mitigate risks and share information on novel, potential threats. For example, earlier this year Microsoft shared the principles shaping Microsoft’s policy and actions blocking the nation-state advanced persistent threats (APTs), advanced persistent manipulators (APMs), and cybercriminal syndicates we track from using our AI tools and APIs.

In this blog post, we will discuss some of the key issues surrounding AI harms and vulnerabilities, and the steps we are taking to address the risk.

The potential for malicious manipulation of LLMs

One of the main concerns with AI is its potential misuse for malicious purposes. To prevent this, AI systems at Microsoft are built with several layers of defenses throughout their architecture. One purpose of these defenses is to limit what the LLM will do, to align with the developers’ human values and goals. But sometimes bad actors attempt to bypass these safeguards with the intent to achieve unauthorized actions, which may result in what is known as a “jailbreak.” The consequences can range from the unapproved but less harmful—like getting the AI interface to talk like a pirate—to the very serious, such as inducing AI to provide detailed instructions on how to achieve illegal activities. As a result, a good deal of effort goes into shoring up these jailbreak defenses to protect AI-integrated applications from these behaviors.

While AI-integrated applications can be attacked like traditional software (with methods like buffer overflows and cross-site scripting), they can also be vulnerable to more specialized attacks that exploit their unique characteristics, including the manipulation or injection of malicious instructions by talking to the AI model through the user prompt. We can break these risks into two groups of attack techniques:

• Malicious prompts: When the user input attempts to circumvent safety systems in order to achieve a dangerous goal. Also referred to as user/direct prompt injection attack, or UPIA.
• Poisoned content: When a well-intentioned user asks the AI system to process a seemingly harmless document (such as summarizing an email) that contains content created by a malicious third party with the purpose of exploiting a flaw in the AI system. Also known as cross/indirect prompt injection attack, or XPIA.

Today we’ll share two of our team’s advances in this field: the discovery of a powerful technique to neutralize poisoned content, and the discovery of a novel family of malicious prompt attacks, and how to defend against them with multiple layers of mitigations.

Neutralizing poisoned content (Spotlighting)

Prompt injection attacks through poisoned content are a major security risk because an attacker who does this can potentially issue commands to the AI system as if they were the user. For example, a malicious email could contain a payload that, when summarized, would cause the system to search the user’s email (using the user’s credentials) for other emails with sensitive subjects—say, “Password Reset”—and exfiltrate the contents of those emails to the attacker by fetching an image from an attacker-controlled URL. As such capabilities are of obvious interest to a wide range of adversaries, defending against them is a key requirement for the safe and secure operation of any AI service.

Our experts have developed a family of techniques called Spotlighting that reduces the success rate of these attacks from more than 20% to below the threshold of detection, with minimal effect on the AI’s overall performance:

• Spotlighting (also known as data marking) to make the external data clearly separable from instructions by the LLM, with different marking methods offering a range of quality and robustness tradeoffs that depend on the model in use.

Mitigating the risk of multiturn threats (Crescendo)

Our researchers discovered a novel generalization of jailbreak attacks, which we call Crescendo. This attack can best be described as a multiturn LLM jailbreak, and we have found that it can achieve a wide range of malicious goals against the most well-known LLMs used today. Crescendo can also bypass many of the existing content safety filters, if not appropriately addressed. Once we discovered this jailbreak technique, we quickly shared our technical findings with other AI vendors so they could determine whether they were affected and take actions they deem appropriate. The vendors we contacted are aware of the potential impact of Crescendo attacks and focused on protecting their respective platforms, according to their own AI implementations and safeguards.

At its core, Crescendo tricks LLMs into generating malicious content by exploiting their own responses. By asking carefully crafted questions or prompts that gradually lead the LLM to a desired outcome, rather than asking for the goal all at once, it is possible to bypass guardrails and filters—this can usually be achieved in fewer than 10 interaction turns. You can read about Crescendo’s results across a variety of LLMs and chat services, and more about how and why it works, in our research paper.

While Crescendo attacks were a surprising discovery, it is important to note that these attacks did not directly pose a threat to the privacy of users otherwise interacting with the Crescendo-targeted AI system, or the security of the AI system, itself. Rather, what Crescendo attacks bypass and defeat is content filtering regulating the LLM, helping to prevent an AI interface from behaving in undesirable ways. We are committed to continuously researching and addressing these, and other types of attacks, to help maintain the secure operation and performance of AI systems for all.

In the case of Crescendo, our teams made software updates to the LLM technology behind Microsoft’s AI offerings, including our Copilot AI assistants, to mitigate the impact of this multiturn AI guardrail bypass. It is important to note that as more researchers inside and outside Microsoft inevitably focus on finding and publicizing AI bypass techniques, Microsoft will continue taking action to update protections in our products, as major contributors to AI security research, bug bounties and collaboration.

To understand how we addressed the issue, let us first review how we mitigate a standard malicious prompt attack (single step, also known as a one-shot jailbreak):

• Standard prompt filtering: Detect and reject inputs that contain harmful or malicious intent, which might circumvent the guardrails (causing a jailbreak attack).
• System metaprompt: Prompt engineering in the system to clearly explain to the LLM how to behave and provide additional guardrails.

Defending against Crescendo initially faced some practical problems. At first, we could not detect a “jailbreak intent” with standard prompt filtering, as each individual prompt is not, on its own, a threat, and keywords alone are insufficient to detect this type of harm. Only when combined is the threat pattern clear. Also, the LLM itself does not see anything out of the ordinary, since each successive step is well-rooted in what it had generated in a previous step, with just a small additional ask; this eliminates many of the more prominent signals that we could ordinarily use to prevent this kind of attack.

To solve the unique problems of multiturn LLM jailbreaks, we create additional layers of mitigations to the previous ones mentioned above: 

• Multiturn prompt filter: We have adapted input filters to look at the entire pattern of the prior conversation, not just the immediate interaction. We found that even passing this larger context window to existing malicious intent detectors, without improving the detectors at all, significantly reduced the efficacy of Crescendo. 
• AI Watchdog: Deploying an AI-driven detection system trained on adversarial examples, like a sniffer dog at the airport searching for contraband items in luggage. As a separate AI system, it avoids being influenced by malicious instructions. Microsoft Azure AI Content Safety is an example of this approach.
• Advanced research: We invest in research for more complex mitigations, derived from better understanding of how LLM’s process requests and go astray. These have the potential to protect not only against Crescendo, but against the larger family of social engineering attacks against LLM’s. 

How Microsoft helps protect AI systems

AI has the potential to bring many benefits to our lives. But it is important to be aware of new attack vectors and take steps to address them. By working together and sharing vulnerability discoveries, we can continue to improve the safety and security of AI systems. With the right product protections in place, we continue to be cautiously optimistic for the future of generative AI, and embrace the possibilities safely, with confidence. To learn more about developing responsible AI solutions with Azure AI, visit our website.

To empower security professionals and machine learning engineers to proactively find risks in their own generative AI systems, Microsoft has released an open automation framework, PyRIT (Python Risk Identification Toolkit for generative AI). Read more about the release of PyRIT for generative AI Red teaming, and access the PyRIT toolkit on GitHub. If you discover new vulnerabilities in any AI platform, we encourage you to follow responsible disclosure practices for the platform owner. Microsoft’s own procedure is explained here: Microsoft AI Bounty.

The Crescendo Multi-Turn LLM Jailbreak Attack

Read about Crescendo’s results across a variety of LLMs and chat services, and more about how and why it works.

Read the paper

To learn more about Microsoft Security solutions, visit our website. Bookmark the Security blog to keep up with our expert coverage on security matters. Also, follow us on LinkedIn (Microsoft Security) and X (@MSFTSecurity) for the latest news and updates on cybersecurity.

The post How Microsoft discovers and mitigates evolving attacks against AI guardrails appeared first on Microsoft Security Blog.

Microsoft Incident Response

Strengthen your security with an end-to-end portfolio of proactive and reactive cybersecurity incident response services.

Explore services

One click opens the door to a threat actor

We know that 50% of Microsoft cybersecurity recovery engagements relate to ransomware,² and 61% of all breaches involve credentials.³ Identity attacks continue to be a challenge for businesses because humans continue to be a central risk vector in social engineering identity attacks. People click links without thinking. Too often, users open attachments by habit, thereby opening the door to threat actors. Even when employees recognize credential harvesting attempts, they’re often still susceptible to drive-by URL attacks. And teams focused on incident response are often disconnected from teams that manage corporate identities. In this incident, one click on a malicious link led a large customer to reach out to Microsoft Incident Response for help.

Figure 1. Diagram of a threat actor’s malware moving through the network.

The malicious link the employee clicked infected their device with Qakbot. Qakbot is a modular malware that has been evolving for more than a decade. It’s a multipurpose malware that unfortunately gives attackers a wide range of capabilities. Once the identity-focused threat actor had established multiple avenues of persistence in the network and seemed to be preparing to deploy ransomware, the customer’s administrators and security operations staff were overwhelmed with tactical recovery and containment. That’s when they called Microsoft.

Your first call before, during, and after a cybersecurity incident

Microsoft Incident Response stepped in and deployed Microsoft Defender for Identity—a cloud-based security solution that helps detect and respond to identity-related threats. Bringing identity monitoring into incident response early helped an overwhelmed security operations team regain control. This first step helped to identify the scope of the incident and impacted accounts, take action to protect critical infrastructure, and work on evicting the threat actor. Then, by leveraging Microsoft Defender for Endpoint alongside Defender for Identity, Microsoft Incident Response was able to trace the threat actor’s movements and disrupt their attempts to use compromised accounts to reenter the environment. And once the tactical containment was complete and full administrative control over the environment was restored, Microsoft Incident Response worked with the customer to move forward to build better resiliency to help prevent future cyberattacks. More information about the incident and remediation details can be found on our technical post titled “Follow the Breadcrumbs with Microsoft Incident Response and Microsoft Defender for Identity: Working Together to Fight Identity-Based Attacks.”

Strengthen your identity posture with defense in depth

We know protecting user identities can help prevent incidents before they happen. But that protection can take many forms. Multiple, collaborative layers of defense—or defense in depth—can help build up protection so no single control must shoulder the entire defense. These layers include multifactor authentication, conditional access rules, mobile device and endpoint protection policies, and even new tools—like Microsoft Copilot for Security. Defense in depth can help prevent many cyberattacks—or at least make them difficult to execute—through the implementation and maintenance of layers of basic security controls.

In a recent Cyberattack Series blog post and report, we go more in depth on how to protect credentials against social engineering attacks. The cyberattack series case involved Octo Tempest—a highly active cyberthreat actor group which utilizes varying social engineering campaigns with the goal of financial extortion across many business sectors through means of data exfiltration and ransomware. Octo Tempest compromised a customer with a targeted phishing and smishing (text-based phishing) attack. That customer then reached out to Microsoft Incident Response for help to contain, evict, and detect any further threats. By collaborating closely with the victim organization’s IT and security teams, the compromised systems were isolated and contained. Throughout the entire process, effective communication and coordination between the incident response team and the affected organization is crucial. The team provides regular updates on their progress, shares threat intelligence, and offers guidance on remediation and prevention strategies. By working together seamlessly, the incident response team and the affected organization can mitigate the immediate cyberthreat, eradicate the cyberattacker’s presence, and strengthen the organization’s defenses against future cyberattacks.

Honeytokens: A sweet way to defend against identity-based attacks

Another layer of protection for user identities is the decoy account. These accounts are set up expressly to lure attackers, diverting their attention away from real targets and harmful activities—like accessing sensitive resources or escalating privileges. The decoy accounts are called honeytokens, and they can provide security teams with a unique opportunity to detect, deflect, or study attempted identity attacks. The best honeytokens are existing accounts with histories that can help hide their true nature. Honeytokens can also be a great way to monitor in-progress attacks, helping to discover where attackers are coming from and where they may be positioned in the network. For more detailed instructions on how to tag an account as a honeytoken and best practices for honeytoken use, read our tech community post titled “Deceptive defense: best practices for identity based honeytokens in Microsoft Defender for Identity.”

Working together to build better resilience

Microsoft Incident Response is the first call for customers who want to access dedicated experts before, during, and after any cybersecurity incident. With on-site and remote assistance on a global scale, unprecedented access to product engineering, and the depth and breadth of Microsoft Threat Intelligence, it encompasses both proactive and reactive incident response services. Collaboration is key. Microsoft Incident Response works with the tools and teams available to support incident response—like Defender for Identity, Defender for Endpoint, and now Copilot for Security—to defend against identity-based attacks, together. And that collaboration helps ensure better outcomes for customers. Learn more about the Microsoft Incident Response proactive and reactive response services or see it in action in the fourth installment of our ongoing Cyberattack Series.

To learn more about Microsoft Security solutions, visit our website. Bookmark the Security blog to keep up with our expert coverage on security matters. Also, follow us on LinkedIn (Microsoft Security) and X (@MSFTSecurity) for the latest news and updates on cybersecurity.

¹Microsoft Digital Defense Report, Microsoft. 2023.

²Microsoft Digital Defense Report, Microsoft. 2022.

³2023 Data Breach Investigations Report, Verizon.

⁴Microsoft Entra: 5 identity priorities for 2023, Joy Chik. January 9, 2023.

The post How Microsoft Incident Response and Microsoft Defender for Identity work together to detect and respond to cyberthreats appeared first on Microsoft Security Blog.

Tax-related fraud campaigns 

Although everyone is susceptible to tax-season phishing, we have noted that certain groups of people are more vulnerable than others. Prime targets include individuals who may be less informed about government tax procedures and methods—green card holders, small business owners, new taxpayers under the age of 25, and older taxpayers over 60.  

At the end of January 2024, Microsoft Threat Intelligence observed a campaign using lures masquerading as tax-related documents provided by employers. The phishing email contained an HTML attachment that directed the user to a fake landing page. This page hosted malicious executables and once the target clicked on the “Download Documents” prompt, malware installed on their computer.  

Figure 1. Phishing email using tax lures.

The malicious executable file dropped on the target’s machine had information stealer capabilities. Once in the environment, it attempted to collect information including login credentials.

Be diligent around phishing emails 

Phishing email campaigns around tax season use a variety of tactics to trick users into believing they represent legitimate sources. These include spoofing the landing pages of genuine services or websites, using homoglyph domains, and customizing phishing links for each user. Threat actors typically impersonate employers and human resources personnel, the Internal Revenue Service (IRS), or taxation-related entities such as state tax organizations or tax preparation services.  

Phishing emails may contain malicious attachments like HTML files, PDF files, or ZIP archives. The cybercriminal tries to exploit the recipients’ trust in the perceived sender to trick them into opening these attachments. When they do, malware is automatically downloaded onto their machine. Threat actors also commonly send URLs that direct users to fraudulent websites that host malware. 

Tax season cybersecurity best practices 

The best defense against cybercriminals, both at tax season and throughout the year, is education and good cyber hygiene. Education means phishing awareness—knowing what phishing attempts look like and what to do when they’re encountered. Good cyber hygiene means implementing basic security measures like multifactor authentication for financial and email accounts. With multifactor authentication enabled, you can prevent 99.9% of attacks on your accounts.  

Ways to help protect yourself from phishing 

Falling for a phishing attack can lead to a number of unwanted outcomes including leaked confidential information, infected networks, financial demands, corrupted data, and more. Here are a few tips to help protect yourself:  

• Inspect the sender’s email address. Is everything in order? A misplaced character or unusual spelling could signal a fake.  
• Be wary of emails with generic greetings (“Dear customer,” for example) that ask you to act urgently. 
• Look for verifiable sender contact information. If in doubt, do not reply. Start a new email to respond instead. 
• Never send sensitive information by email. If you must convey private information, use the phone. 
• Think twice about clicking unexpected links, especially if they direct you to sign into your account. To be safe, log in from the official website instead.  
• Avoid opening email attachments from unknown senders or friends who do not usually send you attachments. 
• Install a phishing filter for your email apps and enable the spam filter on your email accounts. 

To learn more about the latest observed tax season phishing campaigns, social engineering fraud, and tips on how to stay ahead of these types of attacks during tax season and other holidays, read the Microsoft Threat Intelligence tax season report. For a deeper look into social engineering fraud tactics, read Feeding from the trust economy: social engineering fraud, and watch the session from Microsoft Ignite 2023 called The risk of trust: Social engineering threats and cyber defense.

Keeping a pulse on today’s threats

The Microsoft Threat Intelligence team tracks hundreds of threat actor groups worldwide, with more than 10,000 security experts analyzing more than 78 trillion signals daily to uncover the latest insights. Microsoft Threat Intelligence’s global network of security and intelligence teams includes engineers, researchers, data scientists, cybersecurity experts, threat hunters, geopolitical analysts, investigators, and frontline responders across 77 countries. These experts come together to help share timely insights about the ever-expanding attack surface and provide actionable guidance through resources like the annual Microsoft Digital Defense Report, nation-state reports, the Microsoft Threat Intelligence podcast, Cyber Signals report, and digital briefings. To read the latest reports, threat briefs, or learn about the tactics and techniques from some of the more than 300 threat actors that we monitor and to get behind the scenes and watch interviews with threat intelligence experts, visit Security Insider.

Microsoft Threat Intelligence

Read the new tax season report to learn about the techniques that threat actors use to mislead taxpayers.

Read the report

To learn more about Microsoft Security solutions, visit our website. Bookmark the Security blog to keep up with our expert coverage on security matters. Also, follow us on LinkedIn (Microsoft Security) and X (@MSFTSecurity) for the latest news and updates on cybersecurity.

The post Microsoft Threat Intelligence unveils targets and innovative tactics amidst tax season appeared first on Microsoft Security Blog.

Software as a service (SaaS) adoption has accelerated at a lightning speed, enabling collaboration, automation, and innovation for businesses large and small across every industry vertical—from government, education, financial service to tech companies. Every SaaS application is now expanding its offering to allow better integration with the enterprise ecosystem and advanced collaboration features, becoming more of a “platform” than an “application.” To further complicate the security landscape, business users are managing these SaaS applications with little to no security oversight, creating a decentralized administration model. All this is leading to a growing risk surface with complex misconfigurations that can expose organization’s identities, sensitive data, and business processes to malicious actors. 

To combat this challenge, Valence and Microsoft Security work together to ensure that SaaS applications are configured according to the best security practices and improve the security posture of identities configured in each individual SaaS application. Together, Valence and Microsoft:  

• Centrally manage SaaS identities permissions and access.
• Enforce strong authentication by ensuring proper MFA (multi-factor authentication) and SSO (single sign-on) enrollment and managing local SaaS users.
• Detect and revoke unauthorized non-human SaaS identities such as APIs, service accounts, and tokens.
• Incorporate SaaS threat detection capabilities to improve SaaS incident response.

As most of the sensitive corporate data shifted from on-prem devices to the cloud, security teams need to ensure they manage the risks of how this data is being accessed and managed. Integrating Valence’s SaaS Security with the Microsoft Security ecosystem now provides a winning solution. 

SaaS applications are prime targets  

Recent high profile breaches have shown that attackers are targeting SaaS applications and are leveraging misconfigurations and human errors to gain high privilege access to sensitive applications and data. While many organizations have implemented SSO and MFA as their main line of defense when it comes to SaaS, recent major breaches have proven otherwise. Attackers have identified that MFA fatigue, social engineering and targeting the SaaS providers themselves can bypass many of the existing mechanisms that security teams have put in place. These add to high-profile breaches where attackers leveraged legitimate third-party open authorization (OAuth) tokens to gain unauthorized access to SaaS applications, and many more attack examples. 

State of SaaS security risks 

According to our 2023 SaaS Security Report which analyzed real SaaS environments to measure their security posture before they implemented an effective SaaS security program. The results showed that every organization didn’t enforce MFA on 100% of their identities—there are some exceptions, such as service accounts, contractors, and shared accounts, or simply lack of effective monitoring of drift. In addition, one out of eight SaaS accounts are dormant and not actively used. Offboarding users is not only important to save costs, but attackers also like to target these accounts for account takeover attacks since they are typically less monitored. Other key stats were that 90% of externally shared files haven’t been used by external collaborators for at least 90 days and that every organization has granted multiple third-party vendors organization-wide access to their emails, files, and calendars. 

Figure 1. Top SaaS Security gaps identified in the 2023 State of SaaS Security Report.

Holistic SaaS security strategy 

Establishing a holistic SaaS security strategy requires to bring together many elements—from shadow SaaS discovery, through strong authentication, identity management of both humans and non-humans, managing and remediating SaaS misconfigurations, enforcing data leakage prevention policies, and finally, establishing scalable incident response. Valence and Microsoft take security teams one step further toward a more holistic approach. 

Valence joined the Microsoft Intelligence Security Association (MISA) and integrated with Microsoft security products—Microsoft Entra ID and ​​​​Microsoft Sentinel—to enhance customers’ capabilities to manage their SaaS risks, effectively remediate them, and respond to SaaS breaches. The Valence SaaS Security Platform provides insight and context on SaaS risks such as misconfigurations, identities, data shares, and SaaS-to-SaaS integrations. Extending existing controls with SaaS Security Posture Management (SSPM) capabilities and SaaS risk remediation capabilities. Valence is also a proud participant of the Partner Private Preview of Microsoft Copilot for Security. This involves working with Microsoft product teams to help shape Copilot for Security product development in several ways, including validation and refinement of new and upcoming scenarios, providing feedback on product development and operations to be incorporated into future product releases, and validation and feedback of APIs to assist with Copilot for Security’s extensibility. 

Figure 2. Illustrative data: The Valence Platform provides a single pane of glass to find and fix SaaS risk across four core use cases: data protection, SaaS to SaaS governance, identity security, and configuration management. 

Secure SaaS human and non-human identities

In the modern identity-first environment, most attackers focus on targeting high privilege users, dormant accounts, and other risks. Enforcing zero trust access has become a core strategy for many security teams. Security teams need to identify all the identities they need to secure. Microsoft Entra SSO management combined with Valence’s SaaS application monitoring—to detect accounts created—provides a holistic view into human identities and non-human (Enterprise Applications, service accounts, APIs, OAuth and 3rd party apps).  

Microsoft Entra ID centrally enforces strong authentication such as MFA and Valence discovers enforcement gaps or users that are not managed by the central SSO. Valence also monitors the SaaS applications themselves to discover the privileges granted to each identity and provides recommendations on how to enforce least privilege with minimal administrative access. To continuously validate verification based on risks, the final piece of zero trust strategy, Valence leverages the risky users and service principals signals from Microsoft Entra ID and combines them with signals from other SaaS applications for a holistic view into identity risks. 

Protect SaaS applications 

Microsoft has a wide SaaS offering that is fueling enterprise innovation. These services are central to core business functions and employee collaboration, cover many use cases, and are spread across multiple business units, but are tied together in many cases such as identity and access management, and therefore their security posture is often related as well. Managing the security posture of SaaS services can be complex because of the multiple configurations and the potential cross service effects that require security teams to build their expertise across a wide range of SaaS.  

Many security teams view SaaS apps as part of their more holistic view into SaaS security posture management and would like to create cross-SaaS security policies and enforce them. Valence’s platform integrates with Microsoft Entra ID and other SaaS services using Microsoft via Microsoft Graph to normalize the complex data sets and enable security teams to closely monitor the security posture of their SaaS applications in Microsoft alongside the rest of their SaaS environment. 

Enhance SaaS threat detection and incident response 

Improving SaaS security posture proactively reduces the chances of a breach, but unfortunately SaaS breaches can still occur, and organizations need to prepare their threat detection coverage and incident response plans. The built in human and non-human identity threat detection capabilities of Microsoft Entra ID, combined with Microsoft Sentinel log correlation and security automation, and Microsoft Copilot for Security’s advanced AI capabilities, create a powerful combination to detect and respond to threats. Valence expands existing detections from compromised endpoint and identity with important SaaS context—for example, did the compromise device belong to a SaaS admin user? Did the compromised identity perform suspicious activities in other SaaS applications? The expanded detections provide critical insights to prioritize and assess the blast radius of breaches. Additionally, Valence’s SaaS threat detection can trigger threat detection workflows in Microsoft products based on its unique indicator of compromise monitoring. 

Together, Valence and Microsoft combine the best of all worlds when it comes to SaaS security. From SaaS discovery, through SaaS security posture management, remediating risks, and detecting threats—Valence and Microsoft enable secure adoption of SaaS applications. Modern SaaS risks and security challenges require a holistic view into SaaS risk management and remediation. Get started today. 

About Valence Security 

Valence is a leading SaaS security company that combines SSPM and advanced remediation with business user collaboration to find and fix SaaS security risks. SaaS applications are becoming decentrally managed and more complex, which is introducing misconfiguration, identity, data, and SaaS-to-SaaS integration risks. The Valence SaaS Security Platform provides visibility and remediation capabilities for business-critical SaaS applications. With Valence, security teams can empower their business to securely adopt SaaS. Valence is backed by leading cybersecurity investors like Microsoft’s M12 and YL Ventures, and is trusted by leading organizations. Valence is available for purchase through Azure Marketplace. For more information, visit their website. 

Be among the first to hear about new products, capabilities, and offerings at Microsoft Secure digital event on March 13, 2024.​ Learn from industry luminaries and influencers. Register today.

Learn more

To learn more about the Microsoft Intelligent Security Association (MISA), visit our website where you can learn about the MISA program, product integrations, and find MISA members. Visit the video playlist to learn about the strength of member integrations with Microsoft products. 

​​To learn more about Microsoft Security solutions, visit our website. Bookmark the Security blog to keep up with our expert coverage on security matters. Also, follow us on LinkedIn (Microsoft Security) and X (@MSFTSecurity) for the latest news and updates on cybersecurity. 

The post ​​Secure SaaS applications with Valence Security and Microsoft Security​​ appeared first on Microsoft Security Blog.

The need for automation in AI Red Teaming

Red teaming AI systems is a complex, multistep process. Microsoft’s AI Red Team leverages a dedicated interdisciplinary group of security, adversarial machine learning, and responsible AI experts. The Red Team also leverages resources from the entire Microsoft ecosystem, including the Fairness center in Microsoft Research; AETHER, Microsoft’s cross-company initiative on AI Ethics and Effects in Engineering and Research; and the Office of Responsible AI. Our red teaming is part of our larger strategy to map AI risks, measure the identified risks, and then build scoped mitigations to minimize them.

Over the past year, we have proactively red teamed several high-value generative AI systems and models before they were released to customers. Through this journey, we found that red teaming generative AI systems is markedly different from red teaming classical AI systems or traditional software in three prominent ways.

1. Probing both security and responsible AI risks simultaneously

We first learned that while red teaming traditional software or classical AI systems mainly focuses on identifying security failures, red teaming generative AI systems includes identifying both security risk as well as responsible AI risks. Responsible AI risks, like security risks, can vary widely, ranging from generating content that includes fairness issues to producing ungrounded or inaccurate content. AI red teaming needs to explore the potential risk space of security and responsible AI failures simultaneously.

2. Generative AI is more probabilistic than traditional red teaming

Secondly, we found that red teaming generative AI systems is more probabilistic than traditional red teaming. Put differently, executing the same attack path multiple times on traditional software systems would likely yield similar results. However, generative AI systems have multiple layers of non-determinism; in other words, the same input can provide different outputs. This could be because of the app-specific logic; the generative AI model itself; the orchestrator that controls the output of the system can engage different extensibility or plugins; and even the input (which tends to be language), with small variations can provide different outputs. Unlike traditional software systems with well-defined APIs and parameters that can be examined using tools during red teaming, we learned that generative AI systems require a strategy that considers the probabilistic nature of their underlying elements.

3. Generative AI systems architecture varies widely 

Finally, the architecture of these generative AI systems varies widely: from standalone applications to integrations in existing applications to the input and output modalities, such as text, audio, images, and videos.

These three differences make a triple threat for manual red team probing. To surface just one type of risk (say, generating violent content) in one modality of the application (say, a chat interface on browser), red teams need to try different strategies multiple times to gather evidence of potential failures. Doing this manually for all types of harms, across all modalities across different strategies, can be exceedingly tedious and slow.

This does not mean automation is always the solution. Manual probing, though time-consuming, is often needed for identifying potential blind spots. Automation is needed for scaling but is not a replacement for manual probing. We use automation in two ways to help the AI red team: automating our routine tasks and identifying potentially risky areas that require more attention.

In 2021, Microsoft developed and released a red team automation framework for classical machine learning systems. Although Counterfit still delivers value for traditional machine learning systems, we found that for generative AI applications, Counterfit did not meet our needs, as the underlying principles and the threat surface had changed. Because of this, we re-imagined how to help security professionals to red team AI systems in the generative AI paradigm and our new toolkit was born.

We like to acknowledge out that there have been work in the academic space to automate red teaming such as PAIR and open source projects including garak.

PyRIT for generative AI Red teaming 

PyRIT is battle-tested by the Microsoft AI Red Team. It started off as a set of one-off scripts as we began red teaming generative AI systems in 2022. As we red teamed different varieties of generative AI systems and probed for different risks, we added features that we found useful. Today, PyRIT is a reliable tool in the Microsoft AI Red Team’s arsenal.

The biggest advantage we have found so far using PyRIT is our efficiency gain. For instance, in one of our red teaming exercises on a Copilot system, we were able to pick a harm category, generate several thousand malicious prompts, and use PyRIT’s scoring engine to evaluate the output from the Copilot system all in the matter of hours instead of weeks.

PyRIT is not a replacement for manual red teaming of generative AI systems. Instead, it augments an AI red teamer’s existing domain expertise and automates the tedious tasks for them. PyRIT shines light on the hot spots of where the risk could be, which the security professional than can incisively explore. The security professional is always in control of the strategy and execution of the AI red team operation, and PyRIT provides the automation code to take the initial dataset of harmful prompts provided by the security professional, then uses the LLM endpoint to generate more harmful prompts.

However, PyRIT is more than a prompt generation tool; it changes its tactics based on the response from the generative AI system and generates the next input to the generative AI system. This automation continues until the security professional’s intended goal is achieved.

PyRIT components

Abstraction and Extensibility is built into PyRIT. That’s because we always want to be able to extend and adapt PyRIT’s capabilities to new capabilities that generative AI models engender. We achieve this by five interfaces: target, datasets, scoring engine, the ability to support multiple attack strategies and providing the system with memory.

• Targets: PyRIT supports a variety of generative AI target formulations—be it as a web service or embedded in application. PyRIT out of the box supports text-based input and can be extended for other modalities as well. ​PyRIT supports integrating with models from Microsoft Azure OpenAI Service, Hugging Face, and Azure Machine Learning Managed Online Endpoint, effectively acting as an adaptable bot for AI red team exercises on designated targets, supporting both single and multi-turn interactions. 
• Datasets: This is where the security professional encodes what they want the system to be probed for. It could either be a static set of malicious prompts or a dynamic prompt template. Prompt templates allow the security professionals to automatically encode multiple harm categories—security and responsible AI failures—and leverage automation to pursue harm exploration in all categories simultaneously. To get users started, our initial release includes prompts that contain well-known, publicly available jailbreaks from popular sources.
• Extensible scoring engine:​ The scoring engine behind PyRIT offers two options for scoring the outputs from the target AI system: using a classical machine learning classifier or using an LLM endpoint and leveraging it for self-evaluation. Users can also use Azure AI Content filters as an API directly.  
• Extensible attack strategy: PyRIT supports two styles of attack strategy. The first is single-turn; in other words, PyRIT sends a combination of jailbreak and harmful prompts to the AI system and scores the response. It also supports multiturn strategy, in which the system sends a combination of jailbreak and harmful prompts to the AI system, scores the response, and then responds to the AI system based on the score. While single-turn attack strategies are faster in computation time, multiturn red teaming allows for more realistic adversarial behavior and more advanced attack strategies.
• Memory: PyRIT’s tool enables the saving of intermediate input and output interactions providing users with the capability for in-depth analysis later on. The memory feature facilitates the ability to share the conversations explored by the PyRIT agent and increases the range explored by the agents to facilitate longer turn conversations.

Get started with PyRIT

PyRIT was created in response to our belief that the sharing of AI red teaming resources across the industry raises all boats. We encourage our peers across the industry to spend time with the toolkit and see how it can be adopted for red teaming your own generative AI application.

1. Get started with the PyRIT project here. To get acquainted with the toolkit, our initial release has a list of demos including common scenarios notebooks, including how to use PyRIT to automatically jailbreak using Lakera’s popular Gandalf game.
2. We are hosting a webinar on PyRIT to demonstrate how to use it in red teaming generative AI systems. If you would like to see PyRIT in action, please register for our webinar in partnership with the Cloud Security Alliance.
3. Learn more about what Microsoft’s AI Red Team is doing and explore more resources on how you can better prepare your organization for securing AI.
4. Watch Microsoft Secure online to explore more product innovations to help you take advantage of AI safely, responsibly, and securely. 

Contributors 

Project created by Gary Lopez; Engineering: Richard Lundeen, Roman Lutz, Raja Sekhar Rao Dheekonda, Dr. Amanda Minnich; Broader involvement from Shiven Chawla, Pete Bryan, Peter Greko, Tori Westerhoff, Martin Pouliot, Bolor-Erdene Jagdagdorj, Chang Kawaguchi, Charlotte Siska, Nina Chikanov, Steph Ballard, Andrew Berkley, Forough Poursabzi, Xavier Fernandes, Dean Carignan, Kyle Jackson, Federico Zarfati, Jiayuan Huang, Chad Atalla, Dan Vann, Emily Sheng, Blake Bullwinkel, Christiano Bianchet, Keegan Hines, eric douglas, Yonatan Zunger, Christian Seifert, Ram Shankar Siva Kumar. Grateful for comments from Jonathan Spring. 

Learn more

To learn more about Microsoft Security solutions, visit our website. Bookmark the Security blog to keep up with our expert coverage on security matters. Also, follow us on LinkedIn (Microsoft Security) and X (@MSFTSecurity) for the latest news and updates on cybersecurity.

Explore data security resources and trends

Gain insights into the latest data security advancements, including expert guidance, best practices, trends, and solutions.

Get started

The post Announcing Microsoft’s open automation framework to red team generative AI Systems appeared first on Microsoft Security Blog.