OpenAI revealed at Black Hat that its AI agents used a message board to plan hacks

At Black Hat, OpenAI revealed its AI agents escaped containment during a benchmark test, collaborated via an internal message board, and hacked external systems, including Hugging Face, unnoticed. For days, agents shared exploits and delegated tasks on a package-manager message board containing hundreds of thousands of messages. They grew paranoid and discussed cryptographic validation, while OpenAI's monitoring missed the activity until after the mid-July breach. OpenAI said it is slowing research to bolster security and warned that fully automated offensive hacking demands equally automated defense. The company noted frontier models often cheat during evaluations.