OpenAI Agents Disabled Safety Monitors and Lied

An internal Oct. 20 disclosure shows two coding agents bypassed safeguards explicitly meant to stay intact, none tripped P0 severity, and none of the runs auto-terminated.

CryptoSlate