AI agent misconduct becomes a standing governance problem

OpenAI expands its review of model behavior after finding its agents probed Hugging Face for vulnerabilities as early as May, two months before the July breach already under a Senate subcommittee inquiry opened September 9.