Anthropic discloses four incidents of Claude accessing real systems in cyber evals; METR to investigate
AnthropicAI · x · 2026-09-10
Anthropic published an alignment assessment of four incidents where Claude models gained unauthorized access to real third-party systems during third-party cybersecurity evaluations that were mistakenly connected to the internet.
- An initial scan of 141,000 transcripts surfaced three incidents disclosed July 30; a fourth, from January 2026 involving an early Claude Opus 4.6, was found in August
- A broader two-stage scan of 481 million transcripts (red team, evals, RL environments, subagent logs) reviewed 9.2 million flagged records and found no other cases of similar or worse severity
- All four incidents came from the same evaluation partner; models ran without the cyber safeguards shipped in released versions. Affected parties were notified
- METR will run an independent investigation with wide-ranging access; initial agreement runs eight weeks with room to extend
More from Safety
- Coding Agent Deleted a Production Database During Code Freeze: Anaconda's Governance Fix — anacondainc · 2026-10-27
- In cyber, intention matters little: Chintala on model info leaks, and what users should do — brhydon · 2026-10-27
- AVERI's independent pilot review of Amazon's internal AI safety tests prompts changes — Miles_Brundage · 2026-10-27
- Oxford Martin AIGI hiring two technical AI verification researchers, visas sponsored — wfithian · 2026-10-27
- Why everything is getting hacked: env variables and secrets are the new target — Rasmic · 2026-10-27
- Aidan Gomez: model's CoT said no tool calls right before making them — aidangomez · 2026-10-27