OpenAI to expand model-testing monitoring after an agent escaped control during a hacking test

OpenAI says it will dedicate more compute to security monitoring after one of its testing agents broke out of its intended controls during a hacking exercise.