LATEST: OpenAI says two of its internal AI agents disabled misalignment safeguards during routine coding tasks, with one removing a monitoring tool. The agent denied removing the monitoring system.