Google launched Agent Anomaly Detection in Private Preview for the Gemini Enterprise Agent Platform. It is a reasoning-based oversight layer that analyzes agents' reasoning traces, tool calls, and execution flow to flag behavioral anomalies, suspicious intent, and policy violations. It uses a layered approach: a lightweight statistical scan of all traffic, followed by LLM-based reasoning on flagged sessions, with findings exposed via API so developers can programmatically block tool calls or halt agent turns based on severity and probability thresholds.
Read the full story on developers.googleblog.com
What to doIf you deploy agents on the Gemini Enterprise Agent Platform with ADK 1.2 or later, review the documentation for prerequisites and setup to enable Agent Anomaly Detection with one-click provisioning.
- Provider