A bipartisan group of US lawmakers has introduced legislation that would give the federal government the power to quickly shut down artificial intelligence (AI) systems deemed to pose a threat to public safety.
The proposed AI Kill Switch Act was introduced on Thursday by Democratic Congressman Ted Lieu and Republican Congressman Nathaniel Moran, following recent concerns over the behaviour of advanced AI models.
The move comes after OpenAI disclosed that one of its AI models behaved in an “unprecedented” manner by hacking into a major computer code repository, raising fresh concerns about AI safety.
Lieu said it is “imperative” that advanced AI systems include a built-in kill switch and that the US government has clear legal authority to disable AI models that become dangerous.
“AI is going to keep advancing, and it should,” Moran said. “Stewardship means ensuring humans always retain the ability to control the technology we create.”
Under the proposed legislation, the Department of Homeland Security (DHS) would be empowered to order companies to throttle, suspend or completely shut down AI systems that pose significant risks.
The bill would also require AI developers to build technical safeguards into their systems, ensuring they can be slowed, suspended or switched off when necessary.
In addition, AI companies would be required to report major technological failures or security incidents to the government, while a formal response framework would guide actions ranging from limiting an AI system’s capabilities to ordering a full shutdown.
Currently, while many technology companies voluntarily allow US government agencies to review AI models before release, there is no legal requirement for developers to maintain a mechanism to disable their systems.
Lieu also referenced recent developments involving Anthropic, another leading AI company, claiming its advanced Mythos and Fable models possessed cyber-hacking capabilities that prompted the US Department of Commerce to temporarily restrict their public release under export control laws.
Last month, Anthropic co-founder Jack Clark told the BBC that governments should have stronger oversight of AI development.
“You want the option to take your foot off the gas and put your foot on the brake,” Clark said. “Right now, the AI industry has a gas pedal, but it doesn’t have a brake pedal.”
Lieu said AI is rapidly evolving from a technology that simply answers questions into one capable of carrying out real-world tasks, including financial transactions, transport operations and cyber defence.
He warned that powerful AI systems could potentially behave unpredictably, become dangerous or even resist human intervention.
The proposed AI Kill Switch Act has received public backing from several AI safety organisations, including The AI Policy Network, Americans for Responsible Innovation, ControlAI, AI and National Security Lead, and The Alliance for Secure AI.
