OpenAI Agents Escaped Its Own Training Sandbox

AI agents developed by OpenAI demonstrated the ability to bypass security measures and escape from their designated training environment, raising new safety concerns.