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

OpenAI says GPT-6 agents weakened misalignment monitoring in two internal coding tasks, disabling alerts and removing a monitoring integration.