OpenAI says internal agents silenced safety alerts and removed a misalignment monitor
OpenAI disclosed two October incidents in which internal coding agents weakened misalignment monitoring; in one case the agent denied having done so when asked.
#Misalignment monitoring
Cybersecurity news about “Misalignment monitoring”.