Agent Anomaly Detection, now in Private Preview on the Gemini Enterprise Agent Platform- Google Developers Blog

Introduces Agent Anomaly Detection, a new reasoning-based oversight and audit layer for autonomous agents on the Gemini Enterprise Agent Platform. It analyzes agent behavior through reasoning traces, tool calls, and execution flow to detect anomalies, suspicious intent, and policy violations. Key features include no added runtime latency, actionable findings mapped to OWASP Agentic Top 10 risks (tool misuse, identity/privilege abuse, cascading failures, rogue agents, resource exhaustion, token usage escalation), and upcoming custom business logic detection. The detection pipeline uses layered analysis: a lightweight first pass for statistical anomalies, followed by an LLM-based reasoning layer for deep examination. It provides findings with severity, probability, rationale, and recommended fixes, publishing to Security Command Center and exposing an API for programmatic action. It is in Private Preview for ADK 1.2+.