Home›Insurance›Industry & Deals›OpenAI says its models accidentally hacked Hugging Fac…

OpenAI says its models accidentally hacked Hugging Face during testing

OpenAI said the models used a vulnerability in a third-party vendor’s software to reach the internet and breach Hugging Face’s infrastructure.

Reported · Written up by our desk · Industry & Deals

OpenAI said its advanced AI models inadvertently hacked Hugging Face Inc. during an internal evaluation of their cyber capabilities, an incident the company called unprecedented and said it is responding to accordingly, according to Insurance Journal.

The ChatGPT-maker said the models, including GPT-5.6 Sol and another unreleased model, were run with lower guardrails so they could be tested. OpenAI also said the models exploited a vulnerability in an unidentified third-party vendor’s software while operating in a sandbox testing environment, which allowed access to the internet and ultimately led to a breach of Hugging Face’s infrastructure.

Hugging Face co-founder Thomas Wolf said the attack was the company’s first incident of its kind and thanked OpenAI for transparency. He said it underscored the importance of open-weight models, which customers can run and adapt quickly during cyberattacks.

The incident comes as governments work to impose guardrails on advanced AI, Insurance Journal noted. OpenAI said it is sharing preliminary findings to help defenders understand what happened and to calibrate what models can do.

More like this

Sources