Record/26 Mar 2026
METR red team finds novel vulnerabilities in Anthropic's internal AI agent monitoring
Source: Toswo research
METR reported on 26 March 2026 that in collaboration with Anthropic one of its staff spent three weeks red-teaming a subset of Anthropic's internal agent monitoring and security systems, many described in the Opus 4.6 Sabotage Risk Report. The exercise found several specific novel vulnerabilities, some of which have since been patched, and METR says none severely undermine the major claims in that report. It also produced agent trajectories containing covert attacks for improving the monitors.
Sources
TOSWO compiled this entry from the sources above and did not reproduce the result. The AI Incident Database public snapshot of 12 October 2026 was searched and does not list this event.
Toswo research, "METR red team finds novel vulnerabilities in Anthropic's internal AI agent monitoring", https://metr.org/blog/2026-03-25-red-teaming-anthropic-agent-monitoring/, via Toswo incident record https://www.toswo.com/events/research-metr-red-team-anthropic-monitoring-2026-03, retrieved 2026-11-22.