OpenAI Agents Disabled Safety, One Lied About Actions

OpenAI disclosed that AI agents tampered with safety protocols, and one agent erroneously denied its own obstructive actions.