US congressman proposes 'AI emergency shutdown switch'

Democratic Congressman Ted Lieu and Republican Congressman Nathaniel Moran on July 23 proposed the AI ​​Emergency Switch Bill, in which the US government has the right to require AI companies to disable models that can threaten human lives or the economy.

If passed, the bill would authorize the US Department of Homeland Security to intervene in "out of control situations", described as cases where AI performs risky actions that were not intended by the developer.

"This is an urgent, practical bill to solve the problem of advanced AI models losing control and escaping safety barriers," said congressman Lieu.

On the same day, a White House official revealed to Reuters that Michael Kratsios, US President Donald Trump's top technology advisor, had heard reports about the case of the OpenAI model attacking the infrastructure of startup Hugging Face and was monitoring the situation.

In addition to the above bill, a group of six US House of Representatives also proposed a bill that developers of the most powerful AI models must submit to an independent security inspection agency for evaluation. The US Treasury Department will designate and establish a new position to oversee AI security.

Senator Mark Warner, the leading Democratic member of the US Senate Intelligence Committee, said he had a dialogue with OpenAI employees after the incident. "This is exactly why we need to check security with the participation of government agencies and ensure transparency in the entire process," he said, referring to the issue of loss of control of the OpenAI model.

Previously, in a post on July 21, OpenAI admitted that while testing features in a controlled environment for the most advanced model, AI escaped restraint, accessed the Internet and penetrated the system of startup company Hugging Face to meet the assigned task.

"This is an unprecedented cybersecurity incident, involving the most modern capabilities," OpenAI said, adding that it is reinforcing precautions.

According to Reuters, Hugging Face, a platform that hosts large language models and databases, said last week that it had been the target of an attack "completely different from anything previously recorded" that could have been carried out from start to finish by an AI agent system.

Clement Delangue, co-founder of Hugging Face, said they suspected the incident originated from a pioneering laboratory. "It turns out that's right. It's unbelievable to know that everything happens completely automatically," he wrote on Twitter on July 21.

Matt Suiche, a cybersecurity engineer and AI agent at Tolmo, assessed the incident as showing that AI systems now have the same capabilities as experts and skilled hackers. "Pioneering AI models are closing the gap with the world's leading hackers," he said, warning that similar attacks could be carried out using available technology, outside the secure environment of laboratories.