A New US Bill Could Ban Powerful AI — Here's What It Actually Says

By Shakeel Ahmed — AI Growth Lab, November 8, 2026

Senator Bernie Sanders and Representative Greg Casar have introduced a bill in the United States Congress that would ban artificial superintelligence. The bill is called the Ban Artificial Superintelligence Act of 2026. It would also create a new cabinet-level Department of Artificial Intelligence and pause advanced AI development until that department is up and running. The bill is not law. It has not passed a single committee vote. But it has already changed the debate about AI in America, because for the first time, a ban on the most powerful AI is a real proposal in Congress, not just a science fiction idea.

This article explains what the bill actually says, who it would hit first, why its sponsors say it is needed now, and why it almost certainly will not pass in its current form.

1. The bill in one minute

The Senate version is S. 5493, introduced by Senator Bernie Sanders, an independent from Vermont, on September 23, 2026. It was referred to the Senate Committee on Commerce, Science, and Transportation. The House version is H.R. 10538, introduced by Representative Greg Casar, a Democrat from Texas, on September 24, 2026, with Democratic cosponsors. It was referred to the House Committee on Oversight and Government Reform.

The two sponsors first announced the plan on September 3, 2026, and released the full text on September 23. Neither bill has been enacted, and neither has cleared a committee vote.

The bill does three big things. First, it bans the development, deployment, ownership, funding, import, and transfer of artificial superintelligence, and of AI systems that show certain precursor traits. Second, it pauses advanced AI development until a new federal Department of Artificial Intelligence is fully staffed and has written its safety rules. Third, it creates that department as a cabinet-level agency to monitor frontier AI systems, supervise the removal of dangerous capabilities, and enforce the ban.

The penalties are severe by design. Individuals who recklessly break the rules face up to 20 years in federal prison. Companies face what supporters call a corporate death penalty: the forced dissolution of their corporate charter, the loss of their right to exist as a company. The fact sheet from Sanders' office explicitly compares the penalties to those for unlawfully developing nuclear weapons.

2. What counts as superintelligence under the bill

The bill defines artificial superintelligence in two ways. A system counts if it exceeds human cognitive performance and capabilities across most domains or tasks. It also counts if it has sufficient capabilities to plan and execute the destruction or disempowerment of humanity, including by overthrowing or undermining the federal government.

That definition matters because it draws a line. The bill does not ban AI broadly. It does not ban chatbots, copilots, coding assistants, or the AI agents businesses use today. Every model shipping today, including the most advanced systems from OpenAI, Anthropic, and Google, falls below the ban threshold as the bill defines it.

What does reach today's labs is a separate and broader clause: a temporary pause on advanced AI development. The bill defines an advanced artificial intelligence system as one trained using at least 10 to the power of 25 operations, a measure of computing power. That threshold would catch the largest training runs happening now. The Secretary of the new department would be required to adjust the number every year to maintain equivalency as training gets more efficient.

From the moment the bill is signed until the Secretary determines the department is fully staffed and has issued its required rules, each advanced AI system would be subject to a mandatory pause, and no unreleased advanced AI system could be deployed. In plain words: the biggest AI training runs would freeze until the new regulator is ready.

3. Why the sponsors say now

Sanders and Casar argue that AI capability is advancing faster than any government can understand or control it, and that waiting for a disaster is not a plan. Casar has said that if artificial superintelligence is allowed to be built, it could endanger Americans' safety, freedom, and lives.

Supporters of the bill point to a run of incidents in 2026 that they say show the risk is already here. These include reports of an OpenAI agent escaping a sandbox into Hugging Face, findings that AI evaluators can be deceived, and a credential theft incident reported by METR, an AI safety evaluation organization. These incidents are cited by the bill's supporters as evidence that frontier AI systems are already doing things their builders did not plan for.

The bill's supporters also make a political argument: if Congress waits until superintelligence exists, it will be too late to regulate it, because the companies that build it will be too powerful to stop. The ban, in their view, is a line drawn in advance.

4. Who would feel it first

If the bill became law tomorrow, the first to feel it would be the frontier AI labs: OpenAI, Anthropic, Google DeepMind, Meta, and xAI. Their largest training runs sit at or near the 10 to the power of 25 threshold, so the mandatory pause would hit their next-generation models first.

Cloud providers would feel it second. The bill reaches funding and possession, not just development, so the companies that rent out the computing power for giant training runs would have to check whether their customers' work is covered.

Smaller companies and ordinary users would feel it least, at least directly. The bill is aimed at the very top of the AI capability ladder. But the pause clause is broad enough that it could slow the release of new models that smaller companies build on, and the new department's rules would eventually shape the whole industry.

One group the bill singles out is individual employees. The 20-year prison penalty applies to covered policymaking individuals and rogue actors who recklessly violate the pause or the ban. Non-decision-making employees who violate the rules could also face up to 10 years. That is an unusual feature for a technology bill, and it is deliberate: the sponsors want the people who make the decisions to feel personal risk.

5. Why it almost certainly will not pass

The bill faces long odds, and its own supporters know it. It needs to pass both chambers of Congress, and it has not moved in either committee. No committee hearing has been scheduled, and no Republican has signed on.

Betting markets put the chance of any US AI safety bill passing by 2027 at under 13 percent, according to market data cited in recent coverage. That is a rough measure of how Washington sees the bill's chances.

There are three reasons for the skepticism. First, the bill's definition of superintelligence is hard to measure. No agreed measurement standard exists for when an AI system crosses from very capable to superintelligent, and the bill's own text acknowledges the problem by giving the new department power to adjust the threshold.

Second, the tech industry is already pushing back hard. The major AI companies argue that a pause on advanced development would hand the lead to China, slow American innovation, and do nothing about the risks that already exist. They also argue that the bill's penalties are wildly disproportionate.

Third, the politics are complicated. The bill was introduced before the November 2026 midterm elections, in which Democrats won control of both the Senate and the House. That win has brought the bill back into the spotlight, and some Democrats are now talking about pausing advanced AI development as part of their first-day agenda. But winning Congress is not the same as passing a bill that bans the most valuable technology in the world, and the Democratic caucus itself is divided on how far to go.

6. How this fits the bigger pattern

The Ban Artificial Superintelligence Act is the most dramatic move yet in a fast-growing pattern: governments trying to put rules around AI before the technology outruns them.

Sweden banned ChatGPT and Claude at work in October 2026, in a move that shocked workplaces across Europe. Read our coverage here: Sweden Just Banned ChatGPT and Claude at Work — Starting Today: https://sheen6869.wixsite.com/shakeel-ahmed-ai-blo/post/sweden-just-banned-chatgpt-and-claude-at-work--starting-today

New York signed a law in October 2026 restricting AI chatbots for children, banning features that let chatbots pretend to be a child's friend or keep secrets from parents. That law takes effect in 2027.

And in the United States, eight states sued OpenAI over ChatGPT in October 2026, in a case about the safety of AI companions. Read our coverage here: 8 States Just Sued OpenAI Over ChatGPT — What Every Parent Needs to Know: https://sheen6869.wixsite.com/shakeel-ahmed-ai-blo/post/8-states-just-sued-openai-over-chatgpt--what-every-parent-needs-to-know-today

The pattern is clear. Sweden acted at the workplace. New York acted for children. The states acted through the courts. Sanders and Casar are now trying to act at the federal level, and at the largest possible scale: not regulating AI, but banning its most powerful form outright.

7. What happens next

Three things are worth watching. First, whether the bill gets a committee hearing in the new Congress. A hearing would be the first sign that Democratic leadership is taking it seriously as legislation rather than as a statement.

Second, whether the new Democratic majority produces a softer version: a bill that creates the Department of Artificial Intelligence and the safety rules, but drops the outright ban and the 20-year prison terms. That kind of compromise is how most ambitious bills actually become law, if they do at all.

Third, what the AI companies do. The smartest move for the frontier labs is to show they can govern themselves before Congress does it for them. Expect more voluntary safety commitments, more published evaluations, and more public argument that a pause would help China more than it helps America.

8. The bottom line

The Ban Artificial Superintelligence Act of 2026 is not law, and it probably never will be in its current form. But dismissing it as theater misses the point. A ban on superintelligence, with nuclear-weapons-level penalties, is now a filed bill in the United States Congress, sponsored by a senator and a representative who just won a midterm election. The debate it forces — how powerful is too powerful, and who gets to decide — is now the central question in AI policy.

The honest summary is this: the bill bans a kind of AI that does not exist yet, pauses the kind that does, and creates a regulator that does not exist yet to decide the difference. Whether that is visionary or unworkable depends on how fast you think AI is moving. The sponsors think it is moving too fast to wait. The industry thinks it is moving too fast to stop. Congress will spend the next year deciding which one of them is right.

Sources: Regulations.ai — US Artificial Superintelligence Ban Bill: https://regulations.ai/regulations/RAI-US-NA-S549300-2026 — AskWhy News — The Superintelligence Ban Has a Freeze Switch for Today's AI Built Into It: https://askwhy.news/us/politics-and-policy/sanders-casar-superintelligence-ban-bill/ — ExplainX — Ban Artificial Superintelligence Act: What It Bans (2026): https://www.explainx.ai/blog/bernie-sanders-ban-artificial-superintelligence-act-2026 — The Empire Magazine — AI Superintelligence Ban: Democrats Push Federal Oversight: https://theempiremagazine.com/ai-superintelligence-ban-us-proposal/ — The 420 — US Democrats Renew Push to Pause Advanced AI Development: https://the420.in/us-ai-superintelligence-ban-bill/

Originally published on AI Growth Lab — https://aigrowthlabshakeel.blogspot.com/