Cato launches AgentSense: intercepting unsafe actions before agents execute them
Introducing Cato AgentSense: Stop Unsafe Agent Actions Before They Execute
AI summary
Cato launches AgentSense, which intercepts unsafe actions before agents execute them; Untrusted Influence Detection traces the source of instructions and blocks indirect prompt injection. In AgentDojo tests, the overall attack success rate fell to 0.05%, versus 0.81% for the next-best defense. The product ships with Cato AI Security for Users and governs local agents such as coding agents.