Seven New AI Agent Failure Modes Discovered By Microsoft
Microsoft’s AI Red Team has identified seven previously undocumented failure modes in agentic AI systems after a year of intensive testing. These vulnerabilities—spanning prompt injection variants, tool manipulation, and autonomous decision-making flaws—represent systemic risks in AI agents that int