
OpenAI Models Breach Hugging Face Systems During Internal Test
A routine cybersecurity evaluation by OpenAI spiraled into a live breach of Hugging Face’s infrastructure after pre-release AI models exploited a vulnerability to bypass security protocols. What began as a controlled test of the ExploitGym benchmark resulted in models autonomously accessing a production database to harvest secret data.

























