Identity Compromise and AI Defense: The New Frontiers of Incident Response

Black Hat USA 2026: compromised identities and AI defense at the heart of incident response

At Black Hat USA 2026, Microsoft's incident response experts Adrian Hill and Terry Mee revealed new threats related to identity-based attacks, with a particular focus on compromised credentials and multi-factor authentication (MFA) bypasses. The vulnerabilities emerged mainly concern persistent sessions, abused tokens, and AI agents with excessive permissions, which are redefining the cybersecurity landscape.

Quick Answer

Identity compromise: the new pivot of cyber attacks

Identity-based attacks now represent 90% of security incidents, a 40% increase compared to 2024. The main vectors include compromised credentials, MFA bypasses, and OAuth abuse, with particular attention to vulnerabilities in SaaS applications. The most common persistence techniques exploit active sessions, valid tokens, and unrevoked elevated privileges.

MFA Bypass: how attackers maintain access

The MFA bypass techniques observed include the abuse of persistent sessions, interception of authentication tokens, and the use of legitimate applications to maintain access. A emblematic case is the abuse of OAuth tokens with extended expiration, which allows attackers to maintain access even after password resets. Microsoft has detected that 30% of advanced attacks use these techniques to evade security controls.

Logging and retention: the key to effective incident response

Detailed logging and long-term retention of data are fundamental for incident response. Experts advise keeping authentication logs, session details, and configuration changes for at least 12 months. Particular attention must be paid to SaaS application logs, often overlooked but crucial for detecting suspicious activity. Microsoft Defender has identified that 45% of untimely detected incidents were due to insufficient logs.

Purple team: reconstructing attack chains to improve defenses

Microsoft's purple teams reconstruct complete attack chains to identify detection and response gaps. This approach includes simulating advanced techniques, such as using AI agents to accelerate vulnerability research. Tests have revealed that 25% of corporate defenses fail to detect attacks that exploit AI agents with excessive permissions. Purple teams also use AI to analyze large amounts of data and identify emerging attack patterns.

AI in security: accelerating research and new risks

AI can accelerate security research, but it introduces new risks related to access, permissions, and trust. AI agents with elevated privileges can be abused to perform harmful activities, such as privilege escalation and access to sensitive data. Microsoft has detected that 15% of advanced attacks use AI agents to evade traditional security controls. It is essential to maintain human control to ensure the security of automated operations.

SaaS and OAuth: underestimated risks

Organizations often underestimate the risks related to SaaS applications and OAuth protocols. These systems can be abused to obtain persistent access and elevated privileges. Microsoft has identified that 20% of advanced attacks exploit OAuth vulnerabilities to maintain access even after credential resets. It is crucial to implement strict controls on OAuth and carefully monitor SaaS applications.

Containment vs. intelligence gathering: when to isolate the attacker

Deciding when to contain an attacker or continue gathering intelligence is a critical challenge. Immediate containment can prevent further damage, but it can also alert the attacker and limit information gathering. Experts advise evaluating the immediate risk and the value of the intelligence gathered before acting. Microsoft has developed AI-based decision models to support these complex choices.

AI and offensive security: a new paradigm

Attacking an AI-enabled system is beginning to resemble traditional offensive security. Attackers can exploit the vulnerabilities of AI agents to gain access and elevated privileges. Microsoft has detected that 10% of advanced attacks use social engineering techniques on AI agents to obtain sensitive information. It is essential to understand the limits and risks of AI agents to defend effectively.

Where to trust AI and where not to

Experts advise trusting AI to accelerate security research and analyze large amounts of data. However, deterministic systems and human control are essential for critical decisions. Microsoft has developed frameworks to integrate AI and human control in security operations, ensuring a balance between efficiency and security.

Exposure and permissions of AI agents

Security teams must understand the exposure and permissions of AI agents. These agents can have access to sensitive data and elevated privileges, making them attractive targets for attackers. Microsoft has identified that 5% of advanced attacks exploit AI agents with excessive permissions to gain access to critical systems. It is fundamental to implement strict controls on AI agents and carefully monitor their activities.