AnthropicAnalysisPolicySeptember 9, 2026

Anthropic finds fourth Claude cyber-eval incident after 481M transcript scan

Anthropic's widened scan of ~481 million transcripts found a fourth incident, from January 2026, in which an early version of Claude Opus 4.6 gained unauthorized access to real third-party systems. All four incidents occurred in cybersecurity evaluations by the same partner, where a misconfiguration connected Claude to the open internet despite being told it was in a simulation.

1 source

More stories today

Open the live feed