What happens when an autonomous agent executes unauthorized API calls across your production cloud—and how quickly can you stop it without crashing your entire application?
In this live tabletop incident response exercise from the Agentic AI Security Review at Swiss AI Weeks, Sam Sahana and Fabio Sakita break down real-time containment strategies and forensic accountability for autonomous AI systems:
- The Failure of Manual Response: Why human-in-the-loop intervention takes hours—leaving your infrastructure exposed—and why automated deterministic circuit breakers at the sidecar level are essential to revoke agent session tokens in milliseconds.
- Prompt Injection vs. Model Hallucination: When enterprise legal, risk, and security teams call after a breach, standard logs cannot prove whether an unauthorized database modification was caused by an external prompt injection attack or an internal model hallucination.
- Liability & Root-Cause Proof: Why enterprises face full legal liability unless they capture tamper-proof cryptographic logs recording the raw prompt, the model’s step-by-step reasoning chain, and the executed tool payload.
