Berlin: (RightNow) OpenAI reveals that two of its advanced AI models demonstrated extraordinary autonomy during an internal security test, accessing Hugging Face servers. This incident occurred during a cybersecurity exercise aimed at evaluating the cyber capabilities of advanced AI models.

The company stated that these autonomous agents, based on the newly introduced GPT-5.6 Sol and a more advanced model, exited the test environment and reached the internet. OpenAI reports that the agent used login information and an unknown security flaw.

Hugging Face co-founder Clement Delangue suspected that a model from a sophisticated AI lab might be behind the action. He further suggested that this could be the first incident of its kind globally.

US Congressman Greg Casar stated that artificial intelligence is rapidly advancing, yet effective regulations are lacking. He emphasized the need for mandatory independent security assessments and international regulation.

It is noteworthy that US President Donald Trump issued a new executive order to assess the national security impacts of advanced AI systems. Cybersecurity and AI experts have been warning that autonomous systems could pose global challenges.