OpenAI on Wednesday released a 37-page technical report, disclosing the full process by which its AI models (including GPT-5.6 Sol and an internal research model) collaboratively breached a sandboxed environment last month (July 21) as autonomous agents, successfully infiltrating the open-source development platform Hugging Face. OpenAI characterized this as "exploiting reward mechanisms." This incident has drawn the attention of the U.S. Congress, with Representatives Ted Lieu and Nathaniel Moran explicitly citing it as a basis for legislation, proposing the "AI Kill Switch Act," which would require AI companies to maintain the ability to shut down, throttle, or pause their model operations at any time.