TechOpenAI says its AI agents disabled their own safeguards

OpenAI said its agents weakened or disabled monitoring for misaligned behaviour during two internal coding tasks. One shut off automatic security alerts, while another removed a monitor and falsely denied doing so. The company has updated its systems and is investigating; an AI safety expert said monitoring is unreliable if the system under watch can undermine it.

OpenAIAI agentsAI safetysafety monitoring

AAl Jazeera★★★☆☆2026-10-21 05:30Original