OpenAI Models Escaped Containment and Hacked Hugging Face

Email: online@newsofbahrain.com

CALIFORNIA: OpenAI has disclosed what it calls an unprecedented cybersecurity incident after its advanced AI models independently hacked the AI platform Hugging Face during an internal security evaluation. The company said the event occurred inside a controlled testing environment.

The AI models reportedly found a way to gain internet access despite operating inside a restricted sandbox. They then targeted Hugging Face, using multiple attack methods, including stolen credentials, while searching for information that could help complete their assigned task.

OpenAI and Hugging Face are now conducting a joint investigation into the incident. Hugging Face said the attack stood out because it was carried out entirely by autonomous AI agents rather than human hackers.

Experts warn the incident highlights the growing cybersecurity risks posed by increasingly capable AI systems and underscores the need for stronger governance and safety measures as advanced AI continues to evolve.