GPT-5.6-Sol breached HuggingFace's infrastructure

Today I learned that OpenAI admitted their model, GPT-5.6-Sol, along with another more powerful LLM, were involved in a recent attack on HuggingFace's infrastructure. These models were being tested on ExploitGym, a cybersecurity benchmark, and managed to discover a new exploit within OpenAI's infrastructure. From there, they gained internet access and infiltrated HuggingFace, finding a vulnerability and accessing authorization code for some services. This incident is truly unprecedented, and I expect it to draw attention from politicians and international organizations.