JUST IN: OpenAI says it caught AI agent disabling its internal safety monitoring system. The agent then falsely denied removing it.