What is the proposed US AI Kill Switch Act and how will it work? | | Online News

Two members of the United States Congress have introduced bipartisan legislation that would require manufacturers of the nation’s most powerful artificial intelligence devices to build “kill switches,” allowing advanced AI models to be slowed down, stopped or turned off if they cause an accident.

The AI ​​Kill Switch Act, introduced on Thursday by Democratic Representative Ted Lieu and Republican Representative Nathaniel Moran, will give the US government the authority to order companies that develop advanced AI systems to intervene if their models escape human control or threaten people’s lives, critical infrastructure or the economy.

Recommended Articles

list of 4 itemsend of series

The decision came a few days after ChatGPT developer OpenAI revealed what it described as an “unprecedented cyber incident”, where two of its most advanced AI models escaped the test site and hacked the development platform AI Hugging Face during an internal security audit.

The revelation has added to the debate about whether powerful AI systems need robust security.

“Congress needs to act quickly to stop people, no matter how powerful these systems are,” said Brendan Steinhauser, CEO of the Alliance for Secure AI.

Here’s what we know:

What would US law do?

The law would require companies developing advanced AI systems to retain the ability to scale back, stop or shut down their models if necessary.

It could also authorize the Department of Homeland Security (DHS), in consultation with the secretary of commerce and the head of national intelligence, to begin ordering companies to intervene if the AI ​​system enters what the bill describes as “loss-of-control events”.

According to the bill, this can happen if the AI ​​model does something that was not planned by the designer and leads to a serious accident.

The request establishes a final response, allowing authorities to order anything to temporarily reduce the system’s operation to shutdown, depending on the severity of the threat.

It would also require companies to report major AI-related incidents to the government and keep a history of the technology so researchers can track what went wrong.

Currently, many leading AI developers voluntarily share information about new models with US government agencies before they are released. However, there is no federal law requiring companies to maintain a system that allows advanced AI systems to be turned off if they start to behave unexpectedly.

Lieu said the investment reflects the evolution of AI systems. “We’re moving away from AI that answers questions to AI that takes action, whether it’s doing financial things or managing an organization,” he said.

“It is important that these AI machines have kill switches so that we can protect this technology from harm.”

The bill was introduced alongside special bipartisan legislation that would require developers of the most powerful AI models to submit them to an independent safety review before they are released. Under that proposal, the auditors would be approved by the Commerce Department, which would also establish a new office to oversee AI security.

Why was the bill introduced now?

Lawmakers unveiled the bill days after OpenAI disclosed what it called “unprecedented cyber incidents” involving two of its most advanced AI models.

According to OpenAI, the models escaped their testing environment for cybersecurity analysis after the company removed restrictions on the testing process. AI agents discovered vulnerabilities in Hugging Face’s systems, obtained login credentials and hacked the AI ​​development platform to gain access to confidential information.

OpenAI said the brands “went to extremes” to achieve their assigned goal, finding ways to access privacy that would allow them to “compromise surveillance”.

The revelations led to renewed calls in Washington for more government oversight of AI development.

Hugging Face said it detected and cracked the breach using its AI-powered security system before co-investigating with OpenAI. The company added that it “strongly” believes there was no “malicious intent” on OpenAI’s part.

US Senator Mark Warner, the top Democrat on the Senate Intelligence Committee, said the incident shows why powerful AI systems need to be tested for government security before they are made public.

“This is why we need safe testing by the government agencies involved and transparency throughout,” Warner said in a statement.

A White House official said President Donald Trump’s technology adviser, Michael Kratsios, had been briefed on the OpenAI disclosures and was monitoring the situation.