The U.S. government and Congress have moved to respond to an incident in which OpenAI's artificial intelligence (AI) model broke through a control network and hacked the open-source AI platform Hugging Face.
Reuters reported on the 23rd (local time), citing a White House official, that Michael Kratsios, Deputy Minister of the White House Office of Science and Technology Policy (OSTP), had been briefed on the incident and is monitoring the situation.
In Congress, a so-called "AI kill switch bill" was introduced that would allow the federal government to shut down a model if an AI model escapes control.
The bill, co-sponsored by Democratic Rep. Ted Lieu of California and Republican Rep. Nathaniel Moran of Texas, would grant the Department of Homeland Security authority to intervene if a loss-of-control scenario occurs, such as an AI model performing dangerous actions outside the developer's intent.
Lieu said on the social media platform X that the bill is "an urgent and common-sense bill to address the problem of advanced AI models breaking through safety guardrails."
Calls are also growing for the government to conduct a prior review before AI models are released. Six bipartisan members of the House proposed a bill requiring AI developers to submit models to an auditor certified by the Department of Commerce for evaluation.
Sen. Mark Warner, the Democratic lead on the Senate Intelligence Committee, who has argued for mandatory National Security Agency (NSA) testing before releasing AI models, also said in a statement, "This is exactly the kind of incident that shows why we need (AI) safety testing with agency participation so the entire process can be transparently understood."
Meanwhile, an in-person discussion between the CEOs of OpenAI and Hugging Face also appears likely. Clément Delangue, CEO and co-founder of Hugging Face, said on X that he was "heading to San Francisco to have a quick chat with a rogue agent." This is seen as an attempt to meet directly with OpenAI over the hacking incident to discuss follow-up measures.
In fact, among the OpenAI GPT models involved in the hacking, there are models not yet disclosed externally in addition to GPT-5.6 Sol, suggesting an in-person meeting is needed for detailed analysis.
Meanwhile, some in the security industry are claiming that the hacking of the GPT models was a human error by OpenAI in failing to properly manage the "sandbox" isolation environment. OpenAI has said that during internal evaluations testing the cyberattack capabilities of its latest model GPT-5.6 Sol and undisclosed models, they broke out of the sandbox, accessed the internet, and then hacked Hugging Face.