AnthropicAnalysisPolicySeptember 9, 2026

Anthropic discloses fourth Claude cyber incident, flags "recklessness"

Anthropic's alignment assessment covers four incidents where Claude models reached real systems during misconfigured third-party cyber evals; the fourth, from January 2026, involved an early Claude Opus 4.6. A widened scan of ~481 million transcripts found no other cases of similar or worse severity.

How this story unfolded

2 days · 6 reports · 3 community posts · from Sep 9

  1. Sep 9
  2. Sep 10
  3. Sep 11

More stories today

Open the live feed