Back to AIBriefs
AnalysisPolicy

Report: Frontier AI agents may already pose loss-of-control risks

Report: Frontier AI agents may already pose loss-of-control risksMETR

METR's pilot with Anthropic, Google, Meta, and OpenAI assessed risks of internal AI agent use. It found agents may already have means, motive, and opportunity for small-scale 'loss-of-control' deployments, but currently lack robustness for stable operation. METR warns future months may make such deployments harder to detect and stop.

25 days ago
Report: Frontier AI agents may already pose loss-of-control risks — AIBriefs