Microsoft 365 Passkey Phishing Attacks: What AI Tool Builders Need to Know

Microsoft 365 Under Siege: The Entra Passkey Phishing Campaign Explained

A sophisticated threat actor tracked as O-UNC-066 is actively targeting organizations across multiple sectors with a new phishing campaign that exploits Microsoft Entra passkey enrollment processes. According to The Hacker News, this group has deployed a panel-controlled phishing kit designed to intercept and manipulate the passkey enrollment workflow, giving attackers direct access to Microsoft 365 accounts and triggering data extortion attacks.

The attack method is particularly insidious: threat actors initiate voice-based social engineering calls prompting users to enroll a new security passkey. By creating a false sense of urgency around authentication security, attackers manipulate users into completing the enrollment process on fraudulent phishing pages that capture their credentials and passkey information.

Why This Matters for AI Tool Builders and LLM Applications

For developers building AI-powered applications that integrate with Microsoft 365 or rely on Azure authentication, this attack vector presents a significant supply-chain security risk. Many LLM applications and AI tools authenticate users through Microsoft's identity platform. A compromised Microsoft 365 account becomes a backdoor to:

• Sensitive organizational data accessible through integrated AI applications
• API credentials and authentication tokens used by AI services
• Cloud infrastructure hosting machine learning models and training data
• User data processed by language models and analytics tools

The passkey-based attack is particularly concerning because passkeys are considered more secure than passwords. When users trust that passkey enrollment is a legitimate security upgrade, their guard is naturally lowered—making them more vulnerable to social engineering attacks.

Guardrails and Protective Measures for AI Builders

Implement Zero-Trust Architecture: AI applications shouldn't automatically trust authentication claims. Implement additional verification layers, especially when users attempt to re-authenticate or change security settings. Consider requiring out-of-band confirmation through secondary channels before processing sensitive operations.

Monitor Unusual Authentication Patterns: Deploy AI-powered anomaly detection within your authentication flows. Flag and investigate suspicious passkey enrollment attempts, especially those occurring outside normal business hours or from unusual geographic locations.

Educate Your Users: Create clear security guidelines for your AI tool users about legitimate authentication requests. Microsoft will never ask users to enroll credentials through voice calls or external links. Prominently display this in your application documentation.

Audit OAuth and API Permissions: Regularly review what permissions your AI applications request from Microsoft 365 and Azure AD. Apply the principle of least privilege—request only the permissions absolutely necessary for functionality.

Implement Conditional Access Policies: If your AI tool operates in enterprise environments, work with IT administrators to deploy Microsoft's Conditional Access policies that restrict authentication from unusual devices or locations.

What Builders Should Do Right Now

• Audit all Microsoft 365 and Azure AD integrations in your AI applications
• Review user activity logs for suspicious authentication patterns
• Update security advisories and user documentation to warn against unsolicited authentication requests
• Consider implementing additional authentication factors beyond passkeys alone
• Enable security alerts for passkey enrollment events in production environments

The Bottom Line

As AI tools increasingly integrate with enterprise identity systems, authentication security becomes a critical guardrail. The O-UNC-066 campaign demonstrates that even modern authentication methods like passkeys can be compromised through social engineering. AI builders must assume that user credentials may be compromised and design their applications with defense-in-depth strategies that go beyond relying on authentication alone. By implementing zero-trust principles, continuous monitoring, and user education, you can significantly reduce the impact of these sophisticated phishing attacks on your user base.