Overview
🔐 Workshop 1: Poisoned Logs: Red-Team and Harden an AI Incident Agent Nirmal N S, Cloud Infrastructure & Dev Ops Engineer Teams are connecting AI agents to logs, alerts and tickets so they can investigate incidents faster, but those sources contain text that outsiders can influence. You’ll start as the attacker: write log payloads that trick a working incident agent, built with ADK and Gemini, into leaking data or calling the wrong tool. Then you’ll switch sides and harden the same agent in layers: untrusted-input handling, scoped tool permissions, human approval for risky actions and audit logging. Finally, the attackers return, and you’ll see which defences held.
Speakers
2Schedule
No schedule available for this event.