US lawmakers Bernie Sanders and Greg Casar have introduced legislation that would temporarily prohibit the development of advanced artificial intelligence systems and impose severe criminal penalties on companies and individuals that violate the proposed ban.
Sanders, an independent senator from Vermont, and Casar, a Democratic representative from Texas, unveiled the Ban Artificial Superintelligence Act on Thursday, arguing that increasingly powerful AI systems require stronger government oversight.
The proposal has already drawn criticism from technology experts who warn that restricting advanced AI development in the United States could allow China and Russia to gain an advantage in the global race for artificial intelligence.
Former Cybersecurity and Infrastructure Security Agency executive director Bridget Bean has also raised concerns about China’s rapid progress in AI, warning that Beijing is narrowing the technological gap with the US.
Bean has highlighted the potential risks posed by what she described as “dystopian” AI models, as well as the possibility of cyber and biological weapons being enabled by advanced systems. She has stressed the need for continued American innovation and stronger protection of advanced semiconductor technology to preserve US leadership.
Under the proposed legislation, the US would temporarily halt the development of advanced AI systems capable of matching or surpassing human cognitive abilities.
The bill would also create a new Cabinet-level federal agency responsible for regulating frontier AI systems and enforcing the restrictions.
Among the most controversial provisions is a proposed “corporate death penalty” for technology companies that attempt to evade the ban. Companies found violating the restrictions could lose their legal authority to operate in the US.
Individuals, including software engineers, researchers and executives, could face up to 20 years in federal prison.
The proposed penalties are modelled on existing federal laws governing the illegal development of nuclear weapons.
The legislation specifically targets AI systems considered capable of posing an existential threat to humanity or undermining government authority. Among the prohibited capabilities would be the ability to resist shutdown commands, launch unauthorised cyberattacks or attempt to overthrow the US government.
The bill also gives the proposed federal AI agency authority to supervise the removal of dangerous software and, in certain circumstances, oversee the physical destruction of hardware and infrastructure used to operate prohibited systems.
Supporters of the legislation have presented it as a necessary response to the risks posed by powerful AI systems and what they describe as excessive influence by major technology companies.
However, critics argue that imposing such sweeping restrictions on American AI developers could weaken the country’s technological and economic position while doing little to prevent competitors overseas from continuing their own research.
Daniel Castro, president of the Information Technology and Innovation Foundation, said a US-only restriction could have serious consequences.
“If the United States stops developing advanced AI, China is unlikely to follow. Surrendering the lead in a potentially transformative technology would carry serious economic and national security costs,” Castro said.
AI expert Gary Marcus also criticised the proposal, arguing that a blanket prohibition would be too broad.
“A permanent, unilateral ban on all research into superhuman AI is too broad,” Marcus said.
He added that the legislation would be “a guarantee” that the US falls behind other countries in AI development and described it as “not the right approach,” although he acknowledged that some elements of the bill were “defensible.”
Industry advocates have similarly questioned whether the proposed penalties would be practical or effective in managing AI-related risks.
Aden Hizkias, associate policy director at Chamber of Progress, a technology industry coalition, described the legislation as an overly aggressive response.
“Knee-jerk policymaking is never a good thing, and I think that’s exactly what this proposal boils down to,” Hizkias said.
He argued that AI could deliver significant benefits if companies take greater responsibility for how the technology is developed and deployed.
“AI is an incredible tool that can improve millions of lives, but as AI skepticism grows, it’s on companies to earn trust and prove they are good neighbors. That means transparent construction agreements, bringing their own power generation, and making sure they’re meeting environmental standards,” he added.
