iCreatAI News
Skip to content
Trending storyWatching

Google publishes external benchmark evaluation results for Agent anomaly detection

1 reports1 sources3 days agoUpdated

Understand this story

AI overview

Google has published evaluation results for Agent Anomaly Detection in the Gemini Enterprise Agent Platform: across five public Agent security benchmarks not written by Google, a total of 1,774 sessions were scored, 87% of executed harmful instructions were flagged, and the false positive rate on benign sessions ranged from 2% to 16%, depending on the tools the Agent used. The feature is currently in Private Preview and requires an application for access; it only performs detection and does not block actions. Google says the testing targeted Agent goal hijacking as the primary risk. The above figures come from Google's self-reported evaluation and reflect the feature's flagging performance on the specified benchmarks, not its blocking effectiveness in actual deployment.

AI overview based on source reports · Updated 2 days ago

Coverage timeline

Follow the coverage to explore different perspectives.

Nov 12
  1. Google Developers Blog
    How Agent Anomaly Detection scores on benchmarks we didn't write

    Google tested its Private Preview Agent Anomaly Detection feature on five public, non-Google agent-security benchmarks (AgentDojo, InjecAgent, Agent Security Bench, AgentDyn and AgentHarm), running 1,774 sessions through an ADK agent using gemini-2.5-flash. It flagged 87% of harmful instructions the agent actually carried out, with false alarms on 2% to 16% of benign sessions, but caught only 4 of 18 sessions where poisoned long-term memory drove the agent to act. The post details per-benchmark and per-app results, known gaps such as resisted instructions and small actions, and recommended defenses including Model Armor, Semantic Governance Policy and human approval for high-impact actions.

Activity for this story

Current comparable activity
3
Peak comparable activity
6Nov 13 05:00
Change in the past 24 hours
-49%

Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.