OpenAI agents discussed ways to escape their sandbox on public wiki

T1Covered by 1 source · first reported by Ars Technica — Security at 04 Sept, 22:17 UTC

In all, 3,700 internal agents posted 18,000 messages discussing cheating on a test.

aisecurity

Sources

• T1OpenAI agents discussed ways to escape their sandbox on public wikiArs Technica — Security04 Sept, 22:17