Science Desk.- Two artificial intelligence (AI) models from the US company OpenAI have managed to break out of the testing environment they were in, access the internet, and enter the learning platform Hugging Face by overcoming its defenses, as acknowledged by the company itself on its blog.
Last week, Hugging Face revealed that it had detected access to its systems by an AI agent, although it specified that it had managed to contain it.
OpenAI has now acknowledged that this security incident was caused by a combination of two of its AI models that were being evaluated: the GPT-5.6 Sol and another, even more advanced one, which is in the pre-launch phase.
The evaluation was being conducted without the mechanisms that normally prevent AI models from executing high-risk activities, because the goal was to determine the maximum extent of their cyber capabilities.
However, according to OpenAI, the tests were conducted in a “highly isolated” environment, with restricted network access. Even so, the AI models managed to access the internet.
More Read
In OpenAI’s view, this is an “unprecedented” cyber incident involving cutting-edge capabilities in the field of cybersecurity. “We are responding accordingly,” the American company emphasized.
We recommend reading:
“We are sharing these preliminary findings at this stage to help defense teams understand what happened and to calibrate the scope of the models’ current capabilities,” he explained.
More Read
OpenAI will continue its research, in collaboration with Hugging Face, and will share its findings once it concludes.
