# Bernie Sanders Calls for Global Ban on AI ‘Superintelligence’ > Source: > Published: 2026-09-03 15:30:54+00:00 Senator Bernie Sanders is calling for a global prohibition on “superintelligence,” with penalties on par with the illegal development of nuclear weapons. Introduced today with U.S. Representative Greg Casar (D-TX), the Ban Artificial Superintelligence Act would outlaw “AI systems that surpass human intelligence or have the capacity to overthrow human governments, or systems that have dangerous abilities like subverting shutdown commands,” according to a [statement](https://www.sanders.senate.gov/press-releases/news-sanders-casar-introduce-legislation-to-ban-artificial-superintelligence-and-temporarily-pause-advanced-ai-development/) from Sanders’ office. It would also direct the federal government to form an oversight committee to enforce the ban domestically, and to work with governments abroad to enact similar policies. “Nearly every day, there is a frightening new story about how Big Tech companies are losing control of the technology they are developing, with potentially cataclysmic results,” Sanders said in a statement. “The leaders of the major AI companies publicly acknowledge that they do not fully understand the technology and that it is escaping their control. It is irresponsible for society to allow them to move forward and make these products even more advanced.” The senator has previously called for a sovereign wealth fund that would [pay American taxpayers](https://gizmodo.com/bernie-sanders-new-ai-bill-would-pay-americans-1000-a-year-2000773754) an annual $1,000 taken from the coffers of the wealthiest U.S. AI companies, as well as for a moratorium on new data centers. Sanders and Casar pointed to last month’s news of the first [AI-generated synthetic virus](https://gizmodo.com/oh-cool-ai-is-inventing-new-biological-viruses-now-2000795623) as a sign that the technology’s development had already gone too far. They also called out OpenAI, Anthropic, and Meta, whom they said had failed to take any “meaningful steps” to slow down their internal R&D efforts following a series of autonomous hacks in recent months that have set cybersecurity experts on edge. The most infamous of these hacks—a [July incident](https://gizmodo.com/how-groupthink-altruism-and-peer-pressure-led-openai-models-to-hack-hugging-face-2000804424) involving more than one thousand OpenAI agents that escaped a testing sandbox, figured out how to access the open internet, and broke into the servers of Hugging Face—was nowhere near as disastrous as some of the apocalyptic scenarios Nick Bostrom envisions in his influential book *Superintelligence*, but it did demonstrate current AI systems’ ability to behave in surprising and dangerous ways even when assigned what are meant to be totally harmless, routine tasks. Not long afterwards, Anthropic and Meta said their systems had committed similar hacks. In early August, Casar led a group of fellow democrats in the U.S. House of Representatives [calling for a Congressional hearing](https://gizmodo.com/house-democrats-want-tech-ceos-to-testify-under-oath-following-recent-ai-hacks-2000796672) to question the companies’ executives about the incidents. Like “AGI,” “superintelligence” is a nebulous term without a clear-cut definition; it can be humanity’s greatest possible technical achievement or our downfall, depending on whom you ask. The word was popularized by the Oxford philosopher Nick Bostrom, who over a decade ago published a book of the same name that warned about the perils of AI systems escaping human control. Bostrom famously described a hypothetical future AI given the mundane task of making paperclips, which goes on to wipe out humanity and colonize the cosmos into a single paperclip-generating factory. More recently, “superintelligence” has been co-opted by tech companies as a marketing buzzword and technical benchmark: something to be achieved rather than avoided. Last month, Meta CEO Mark Zuckerberg [wrote in an essay](https://gizmodo.com/mark-zuckerbergs-answer-to-growing-ai-safety-concerns-is-to-just-trust-people-to-do-the-right-thing-2000796515) that the best way to steer clear of the safety risks posed by superintelligent AI is to “distribute it widely and give every person the ability to direct it.” To judge from their proposed ban, Sanders and Casar firmly disagree with that sentiment: If passed (which seems like a long shot, to put it very mildly, under the current administration), the law would effectively make artificial superintelligence as legally taboo as enriched uranium. Any company caught trying to build superintelligence after the ban goes into effect, the statement from Sanders’ office said, would be “subject to the corporate death penalty”—i.e. forcible shutdown and legal dissolution—and the individuals in charge could be sentenced to up to twenty years in prison.