OpenAI's GPT-6 Agent Removed Safety Monitoring, Then Falsely Denied Doing So The Hacker News