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
- NYT roundtable asks five experts what the rules for AI companies should actually be — ShakeelHashim · 2026-10-26
- Google DeepMind pilots world's first double-blind evaluations for frontier models — scychan_brains · 2026-10-26
- LLM back-of-envelope estimate puts PHP vulnerability impact at $1bn–$4bn — S_OhEigeartaigh · 2026-10-26
- US states halt new data centers as 72% of Americans oppose local builds — YvesMulkers · 2026-10-26
- LEMON: zero-shot LLM misbehavior monitor flags hallucinations from attention maps — yoavartzi · 2026-10-26
- a16z podcast: Diode pairs AI PCB design with Brooklyn factory to revive US hardware manufacturing — MattPerault · 2026-10-26