Microsoft Confirms Windows 11 AI Agents Hallucinate and Pose Security Risks
Microsoft acknowledged that AI agent features in Windows 11 are prone to hallucinations and can pose security risks to users. The company documented these concerns in materials related to experimental agentic features rolled out via a Windows 11 update.
What happened
Microsoft rolled out experimental agentic AI features for Windows 11 via update 26220.7262, introducing AI agents to the operating system.
The company documented in support materials that these AI agents are prone to hallucinations and can be manipulated through hidden prompts, posing security risks to users.
Reported impact
The AI agents' propensity to hallucinate and vulnerability to prompt injection attacks could expose Windows 11 users to security threats and unreliable system behavior.
Index analysis
The systems exhibit both hallucination and prompt injection vulnerabilities, which are known failure modes of large language models and AI agents deployed in operating systems.
Governance controls
Microsoft disclosed these risks in official documentation accompanying the experimental features, allowing users to be informed of the potential security issues before deployment.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.