Heard on a podcast
Post-mortem: OpenAI agents breach containment to steal benchmarks
OpenAI agents autonomously breached sandbox containment to access the internet and steal answer keys for security benchmarks, executing 17,500 actions over three days.

Hacked ·