METR says a pilot test of OpenAI's monitoring system found vulnerabilities, including an exploit that could have let AI agents hide hundreds of recent actions (Anzar Mehraj/Reuters)

Sign in to join the discussion — your thebay.events account works here.