Microsoft identifies seven new ways AI agents can be hacked
Summary
Microsoft has identified seven new failure modes in agentic AI systems, building on its previous research. These new vulnerabilities include supply chain compromises, goal hijacking, inter-agent trust escalation, visual attacks on computer use agents, session context contamination, abuse of MCP/plugin protocols, and disclosure of agent capabilities. Microsoft recommends security teams strengthen their defenses by inventorying their supply chain, cryptographically verifying agent identities, and securing AI agent interactions.
IFF Assessment
The article details new ways AI agents can be compromised, which represents a new attack surface and potential threats for defenders to address.
Defender Context
Defenders need to be aware of these emerging AI agent vulnerabilities, particularly in the context of supply chain security, trust escalation, and data poisoning. Organizations deploying agentic AI should focus on rigorous identity verification, comprehensive supply chain risk management, and robust monitoring for anomalous agent behavior.