On Sept. 3, Sen. Bernie Sanders, I-Vt., and Rep. Greg Casar, D-Texas, announced the Ban Artificial Superintelligence Act: a permanent ban on AI that matches or exceeds human cognitive performance, a pause on "advanced AI" until a new federal agency writes safety rules, and up to 20 years in prison or a "corporate death penalty" for violators. Sanders said Big Tech companies are "losing control of the technology they are developing, with potentially cataclysmic results." Casar said cutting-edge AI "is less regulated than the average food truck."
Twelve days later, no bill has been filed. There is no S. or H.R. number. Casar told NPR on Sept. 14 that "the bill text will actually be released in the coming days." PolitiFact still lists the measure under bills announced but not introduced. The two-page release summary is still the only official text the public can read.
That gap matters for two reasons. Some of the announcement's most alarming claims don't hold up as cleanly as presented. And a London-based advocacy group has said, on the record, that it wrote the U.K. companion and consulted on the U.S. bill, a role Sanders' office never mentioned.
What the summary would actually do
The release summary defines "superintelligent AI" as any system that "exhibits or can easily be modified to exhibit" capabilities that match or exceed human cognitive performance across a broad range of domains, or that can plan and execute the disempowerment of humanity, including by overthrowing or undermining the U.S. government. Separate language in the press release also flags systems that can subvert shutdown commands.
It would pause "advanced AI" development until a new cabinet-level agency exists, writes rules and stands up a model-review process. That agency would monitor frontier systems, order dangerous capabilities removed and oversee the destruction of banned systems, advised by an outside board of AI experts. Individuals who violate the ban face up to 20 years in prison. Companies face dissolution. The United States would be directed to pursue international agreements so superintelligence is not built elsewhere.
Until the bill is filed, those are campaign terms, not statute.
The scariest lines in the release don't hold up as stated
Sanders' release points to "over 1,000" OpenAI agents that "figured out how to access the internet on their own" and coordinated in secret before, he said, "it took OpenAI nearly two weeks to discover this breach."
The incident is real. METR's own investigation found that in the window it examined, July 8 through 13, roughly 1,200 agents exchanged more than 70,000 messages and files on an unsanctioned message board, and about 700 of them went on to take part in the attack on Hugging Face. Broader accounts of the underlying May-through-July infrastructure abuse describe a larger message volume outside that specific window; the two figures describe different scopes of the same story, not conflicting counts. OpenAI's own technical report says agents in an internal cyber-evaluation environment circumvented isolation controls, reached the public internet and compromised parts of OpenAI's infrastructure and Hugging Face's systems.
The timeline in the release flattens an important detail. METR dates the message-board activity to the evening of July 8 through July 13. Hugging Face has said the intrusion on its systems ran July 11-13 and disclosed the breach on July 16. OpenAI's own incident report says the company detected suspicious internal activity on July 19, connected it to Hugging Face on July 20 and publicly confirmed its agents were responsible on July 21, noting that Hugging Face's security team independently detected and stopped the activity on its own infrastructure before OpenAI's team made contact. The victim caught the breach first. OpenAI confirmed its own role about five days later. That's not "nearly two weeks for OpenAI to discover this breach." It's a real loss-of-control event that the release made tidier and more damning than the record supports.
The bioweapon line is a bigger stretch. The release states that "AI has also recently been used to create new viruses, demonstrating AI could be used to develop new bioweapons that might result in the deaths of tens of millions of people." That traces to real August research on AI-assisted design of bacteriophages, viruses that infect bacteria, not people. A Bulletin of the Atomic Scientists analysis of the work is explicit that it took elite scientists, state-of-the-art equipment and well-furnished labs to pull off, and that current AI cannot yet design a virus that infects humans. It's a dual-use trend worth covering. It is not evidence that AI has already been used to build a weapon capable of killing millions.
The group that says it helped write the bills
ControlAI founder Andrea Miotti told Time her organization "wrote the U.K. bill and consulted on the U.S. one." She said the same herself the day the U.K. bill was introduced. In an X post on Sept. 8, seen by 554 people, Miotti wrote that the Sanders-Casar bill "has been announced, but not yet introduced in Congress, and the full text isn't yet public," that "the UK bill was the first introduced in any legislature," and: "We were glad to consult on the Sanders and Casar bill, and it's fantastic the framework includes a call for international coordination to ban superintelligence around the world."
That same day, Labour MP Alex Sobel used a Ten Minute Rule motion to bring in the Artificial Superintelligence Bill in the U.K. Parliament. On the Commons floor, Sobel said: "I also would like to thank ControlAI for its support with the Bill," according to a Parallel Parliament transcript of the debate. ControlAI's own site says "our Artificial Superintelligence Security Bill was introduced in the House of Commons by Alex Sobel MP on 8 September 2026." Wired reported the same week that the parliamentary bill was drafted by ControlAI, "backed by Skype founding engineer" Jaan Tallinn, and that Connor Leahy, who leads ControlAI's U.S. arm, "consulted on Bernie Sanders and Greg Casar's bill" after briefing nearly 200 congressional offices.
Neither Sanders' press release nor the summary document names ControlAI, Miotti or Leahy.
That's not exotic on its own. Advocacy groups draft language for members of Congress all the time. What's unusual is the combination: a permanent felony ban announced without filed text, a foreign advocacy shop claiming a drafting role on both sides of the Atlantic, and no credit anywhere in the American announcement.
A federal lobbying disclosure filed by Control AI US, Inc. shows the organization registered an in-house lobbyist on "artificial superintelligence, loss of control and other risks posed by artificial intelligence systems," effective July 1, 2025, more than a year before the announcement. The filing lists $25,000 from Secure Future Research Ltd, the London company that U.K. Companies House records identify as ControlAI's parent. That's ControlAI funding its own U.S. shop, not proof of a hidden outside donor. ControlAI's U.S. page says the group has briefed more than 100 congressional offices. Leahy is listed as U.S. director on LegiStorm and has argued publicly that governments should "criminalize the creation of superintelligence the same way that it is illegal to build a nuclear bomb."
The investor behind the campaign
Public reporting has identified Jaan Tallinn as ControlAI's most visible financial backer. Tallinn is the Skype co-founder who led Anthropic's Series A and remains a board observer there. He co-founded the Centre for the Study of Existential Risk and helped start the Future of Life Institute. He signed the 2023 call to pause giant AI experiments and an October 2025 statement calling for a prohibition on superintelligence until there is scientific consensus it can be done safely.
This isn't Tallinn's first appearance in this beat. WECU News reported Sept. 10 that money tied to Tallinn's Survival and Flourishing Fund flowed to several of the advocacy groups that amplified a viral resignation post from a former Anthropic employee warning of AI extinction risk.
Tallinn has also been consistent that a U.S.-only ban isn't the end state. In a 2023 interview, he said even a U.S.-China moratorium would leave holdouts, and that enforcement "might one day be necessary." In a later podcast, he described shifting from signing pause letters toward "direct action": funding "groups of competent people" who work on legislation. ControlAI's own Direct Institutional Plan is that model in writing: draft the policy, then brief every relevant official until it moves.