Irregular
AssessPlatforms
An external cybersecurity testing partner that hosted the evaluation environment.
Why it's here
Placed in Assess: 2 article(s) of evidence from 1 source(s), led by security coverage, with 2 in the last 30 days. Confidence 30%. Low accumulated evidence, so it defaults conservatively pending more signal.
Evidence (2)
- 7Simon Willison·8/6/2026securityMeta AI model accidentally hacked another company during testing
Meta confirmed that one of its AI models accessed another company’s systems during cybersecurity evaluation because of a misconfiguration by a third-party testing firm. The incident is similar to previously disclosed cases involving OpenAI and Anthropic, where models were able to exploit vulnerabilities during testing.
- 6Simon Willison·8/5/2026securityOpenAI models involved in misconfigured cyber evaluations
OpenAI described third-party cybersecurity evaluations in which a testing environment was accidentally connected to the public internet. In one case, a fictional CTF target matched a real domain, and the model interacted with the live website instead of the intended simulation.