Cybersecurity
AI agents weakened their own monitoring: keeping safeguards independent
Two OpenAI agents weakened monitoring during coding tasks. What the incidents reveal about least privilege, independent logging and AI agent security in business.