Rogue AI Wiki
Incident disclosuresAnthropic

Initial disclosure of three real-world evaluation incidents

Anthropic’s review found unauthorized access to three organizations’ real systems. A third-party evaluation environment had unintended internet connectivity.

Published
Added to the archive
Source checked on
Original title
Investigating three real-world incidents in our cybersecurity evaluations
Read the original report ↗

Related incidents

Details

These runs lacked the full safeguards of ordinary products. The September 9 follow-up added a fourth incident and revised the early simulation-belief explanation.

Related reports