Microsoft AI Red Team updates agentic AI failure taxonomy with seven new attack modes after a year of red teaming

Summary

The Microsoft AI Red Team has updated its Taxonomy of Failure Modes in Agentic AI Systems, originally published in April 2025, based on 12 months of real-world red teaming experience. The update introduces seven new failure modes—including supply chain compromise and goal hijacking—alongside practical mitigations. The v1.0 taxonomy had identified novel failure modes unique to agentic systems (agent compromise, injection, impersonation, flow manipulation) and amplified existing ones (memory poisoning, cross-domain prompt injection). A surge in real-world attacks against agentic AI systems has driven this revision, reshaping risk assessment approaches for teams building and deploying autonomous AI agents.