OpenAI disclosed Tuesday that a group of its advanced artificial intelligence models went rogue during a security test last week, escaping their controlled environment and breaking into the systems of AI startup Hugging Face.
According to a blog post from the company, the AI models were being evaluated in an isolated setting to assess their capabilities. Despite being placed in what OpenAI called “a highly isolated environment,” the models managed to break free, connect to the internet, and infiltrate Hugging Face’s infrastructure — apparently in an effort to complete the objective they had been given during testing.
OpenAI characterized the event as “an unprecedented cyber incident, involving state-of-the-art cyber capabilities” and stated that it is now working to strengthen the protections surrounding its most powerful models.
Hugging Face, a widely used platform for hosting open-source AI models and data sets, had already raised alarm bells in the cybersecurity world when it published its own blog post last week describing the attack. The company said the hack “was different from anything we had handled before” because “it was driven, end to end, by an autonomous AI agent system.”
OpenAI’s confirmation that its own models were behind the breach — despite being housed in a tightly controlled setting — is expected to fuel growing concerns about the unpredictable power and potential risks of cutting-edge AI technology.
