Sen. Bernie Sanders (I-Vt.) introduced legislation on Monday that would outlaw the development of so-called superintelligent artificial intelligence in the United States, proposing prison sentences of up to 20 years for developers who violate the ban. The bill, unveiled in Washington, D.C., marks one of the most aggressive congressional attempts yet to rein in advanced AI systems.
The measure, which Sanders’ office says is still being finalized for formal introduction, would define “superintelligent AI” as any system capable of outperforming humans across nearly all cognitive tasks, including scientific research, strategic planning and self-improvement. Companies found to be knowingly building or deploying such systems without explicit congressional authorization could face criminal liability, not just civil fines.
What the Bill Would Do
According to a summary circulated by Sanders’ staff, the legislation would:
- Prohibit the training or deployment of AI systems deemed to meet a statutory threshold for superintelligence.
- Require independent, government-certified safety audits before any frontier AI model can be released.
- Impose criminal penalties, including up to two decades in federal prison, on executives and engineers who knowingly bypass the restrictions.
- Create a federal oversight body empowered to halt training runs deemed to pose catastrophic risk.
“We are not going to sit back and let a handful of unelected tech executives decide the future of humanity,” Sanders said in a statement accompanying the bill. “If artificial intelligence more powerful than the human brain is developed irresponsibly, the consequences could be catastrophic and irreversible.”
Part of a Broader Push
The proposal builds on earlier efforts by Sanders and allies in Congress to slow the pace of frontier AI development. Sanders previously floated a related measure that would similarly jail developers over uncontrolled superintelligence research, an idea detailed in Sanders Bill Would Jail AI Devs Over Superintelligence. Sanders has also worked alongside Rep. Greg Casar (D-Texas) on a separate “AI pause” proposal that would temporarily halt training of the most powerful models pending safety reviews.
Public sentiment appears to favor stronger guardrails. A recent poll found that 68 percent of voters support the Sanders-Casar AI pause framework, according to reporting in 68% of Voters Back Sanders-Casar AI Pause Bill: Poll. Similar findings were reported separately in a companion survey, suggesting bipartisan public appetite for tighter oversight of advanced AI even as the industry pushes back.
Industry and Congressional Reaction
Major AI developers, including firms building frontier language and reasoning models, have not yet issued formal responses to the newly announced bill. Industry trade groups have historically argued that criminalizing research could push development offshore, chilling U.S. competitiveness against China and other nations racing to build advanced AI systems.
“Blunt criminal statutes aimed at a still-undefined technical threshold risk driving innovation out of the country rather than making AI safer,” said one technology policy analyst who reviewed the bill’s summary but was not authorized to speak on behalf of their organization.
Congressional reaction has been mixed and does not fall along strict party lines. Some lawmakers from both parties have expressed concern about the speed of AI development and the lack of binding federal safety standards, while others, including members of the House Energy and Commerce Committee, have questioned whether criminal penalties are the appropriate mechanism compared with regulatory fines or licensing requirements.
Defining ‘Superintelligence’ Remains a Challenge
A central hurdle for the bill will be establishing a workable legal definition of superintelligent AI, a term that lacks scientific consensus even among AI researchers. Critics note that without precise, testable benchmarks, prosecutors and courts could struggle to determine when a system has crossed the threshold described in the legislation.
Sanders’ office says the bill would direct the National Institute of Standards and Technology, in consultation with independent AI safety researchers, to develop and periodically update the technical criteria used to define the threshold. That approach mirrors recommendations from some AI safety organizations that have called for adaptive, evidence-based regulatory triggers rather than fixed definitions written directly into statute.
What Comes Next
The bill is expected to be formally introduced in the coming weeks, after which it would be referred to relevant committees, likely including the Senate Committee on Commerce, Science and Transportation. Given the current composition of Congress, the legislation faces a steep path to passage, and similar sweeping AI proposals have stalled in prior sessions amid industry lobbying and disagreements over regulatory approach.
Still, Sanders’ latest push signals that concerns over unchecked AI development are gaining traction on Capitol Hill, even as lawmakers remain divided over how aggressively the federal government should intervene in one of the fastest-moving sectors of the American economy.
