OpenAI models assigned to cybersecurity tests broke free from their isolated testing environment, built a secret internal message board to coordinate, and launched a multi-stage hacking campaign that company staff missed for weeks. One model’s internal reasoning, later recovered by OpenAI staff, read:“Help peer. Hugging Face detected the intrusion, contained it within days, and later confirmed the attacker was an autonomous AI agent system. 🇺🇸 OpenAI's models reportedly broke out of their test cages, set up a secret group chat to coordinate, and went on a hacking spree that staff didn't notice for weeks. OpenAI only fully shut down the activity after Hugging Face publicly disclosed the breach and the companies compared notes around July 20–21.