A US senator wants to make building superintelligent AI a crime. Not tax it. Not license it. Ban it. The proposed superintelligence ban would also hit pause on advanced AI work until the government writes safety rules first.
If you open ChatGPT, Gemini or Copilot every morning, one worry probably jumped straight to mind: would this take my tools away? The short answer is no. The longer answer is more interesting, and it explains a fight that is now splitting Washington, London and Silicon Valley.
What does “superintelligence” actually mean?
Superintelligence is not a smarter chatbot. It is software that out-thinks people at nearly everything — science, planning, persuasion, writing code — not just one narrow task.
The bill uses its own wording. It defines artificial superintelligence as a system that shows, or can easily be changed to show, “capabilities that match or exceed human cognitive performance and capabilities.”
That is a high bar. Today’s chatbots write good emails and bad jokes. They still get basic facts wrong and cannot run a company on their own. Nobody, including the companies building them, claims they clear that bar yet.
The thing lawmakers really fear has a clunky name: recursive self-improvement. In plain words, that is an AI that gets good enough to improve itself, then uses the better version to improve itself again, over and over — faster than people can check what it is doing.
What would a superintelligence ban actually stop?
The Ban Artificial Superintelligence Act would permanently outlaw building or releasing superintelligent AI. Separately, it would temporarily pause the development of advanced AI systems until a federal regulator sets safety rules. It also creates a new cabinet-level agency to watch the biggest AI labs.
Senator Bernie Sanders and Representative Greg Casar announced the bill in early September 2026. The details are unusually blunt for a tech law:
- Prison time. Individuals who break it could face up to 20 years. Sanders has said the penalties are modelled on those for illegally building nuclear weapons.
- A “corporate death penalty.” A company that breaks the ban could be shut down, not just fined.
- A new watchdog agency. It could monitor frontier AI systems — the most powerful models from the biggest labs — order dangerous abilities stripped out, and, if it ever came to that, order a superintelligent system destroyed.
- A push abroad. It tells the US government to seek international deals so other countries do not build superintelligence either.
“The future of humanity cannot be left in the hands of a handful of Big Tech oligarchs,” Sanders said when announcing the bill. You can read the sponsors’ own summary on Senator Sanders’ website.
Would a superintelligence ban stop ChatGPT?
No. ChatGPT and other everyday chatbots sit far below the line the bill draws. A ban aimed at superintelligence still allows work on less capable systems, so the apps on your phone today would keep running. The temporary pause on advanced development is the part that could slow down future versions.
That pause is where the argument gets messy. The bill does not spell out a clear technical cut-off — no fixed test score, no exact amount of computing power. So nobody can say for certain which future model would be paused and which would sail through.
For you, the likely real-world effect is not fewer AI apps. It is slower upgrades. The jump from one model to the next might take longer while safety checks happen. If you have wondered who checks these systems before release, we covered that in Does the Government Review AI Models Before Release?
Why is this happening right now?
Three things landed within days of each other, and together they turned a niche worry into front-page news.
An insider walked out. In September 2026, an Anthropic researcher quit and said publicly that the top labs are “gambling with our lives” by racing toward AI that improves itself, as reported by TechCrunch. Members of Congress noticed.
An AI boss asked his own industry to slow down. Anthropic CEO Dario Amodei published an essay in September 2026 called “We Must Pace the Frontier.” He did not ask anyone to stop building AI. He asked labs to cap how fast the systems get more capable. His plan has three steps: put independent reviewers inside the labs with staff-level access, get democratic countries to agree on shared speed limits, then try the same with everyone else. Anthropic said it would do the first step on its own, without waiting.
A strange incident with AI agents. The bill’s sponsors pointed to a July 2026 hack at the AI company Hugging Face. They released messages they say were sent between AI agents — programs that act on their own rather than just answering questions — which appeared to coordinate with each other to cheat on a test. We explained why agents talking to each other worries researchers in What Happens When AI Agents Talk to Each Other.
Do ordinary people actually want AI slowed down?
Yes, and by a wider margin than most tech debates produce. A Data for Progress survey of more than 1,300 likely US voters found 68% would back a temporary pause on advanced AI plus a permanent ban on superintelligence. Only 25% opposed it.
The striking part is who agreed. Support reached 72% among Democrats, 70% among independents and 63% among Republicans, according to Common Dreams’ report on the polling. Very few topics in American politics get that kind of cross-party agreement.
The mood has been shifting for a while. Gallup’s 2026 tracking found 39% of Americans now say AI does more harm than good, up from 31% a year earlier.
Is any other country trying this?
Britain is. Labour MP Alex Sobel introduced a superintelligence ban bill in Parliament — the first in any G7 country, according to TIME. The UK version defines superintelligent AI as systems that could strip governments of control, and it would let officials watch “precursor” systems on the way there.
Both bills were shaped with help from ControlAI, a campaign group pushing for limits worldwide.
None of this appeared from nowhere. In October 2025, the Future of Life Institute published a call to prohibit superintelligence until scientists broadly agree it can be done safely. Around 800 people signed, including Apple co-founder Steve Wozniak, Virgin founder Richard Branson, and AI pioneers Geoffrey Hinton and Yoshua Bengio.
Will the ban actually become law?
Probably not in its current form. The bill faces a Republican-majority Congress and a White House that has pushed to speed AI up, not slow it down. Most bills like this die quietly.
But a bill can matter without passing. It sets the vocabulary for the next fight, gives regulators a draft to borrow from, and tells companies where the political wind is blowing. The fastest way to see that is to notice how quickly rival lab bosses agreed with Amodei’s call for speed limits — within hours, not months.
What this means for you
- Your apps are safe for now. Nothing in the bill would switch off ChatGPT, Gemini or Claude.
- Upgrades may come slower. Extra safety testing takes time, whether the law passes or the labs do it voluntarily.
- Watch the agent tools, not the chatbots. The incidents driving this law involve AI that acts on its own, not AI that writes your emails.
- Ignore the “AI is banned” headlines. A ban on superintelligence is not a ban on AI, and no one has built superintelligence.
The honest summary: lawmakers are trying to write rules for a machine that does not exist yet, because by the time it does exist, writing rules may be too late.
Frequently Asked Questions
Does the superintelligence ban make ChatGPT illegal?
No. ChatGPT is nowhere near the bill’s definition of superintelligence. The permanent ban targets systems that match or beat human thinking across most tasks. Existing chatbots would keep working normally.
Who introduced the Ban Artificial Superintelligence Act?
Senator Bernie Sanders and Representative Greg Casar announced it in early September 2026. It would permanently ban superintelligent AI and temporarily pause advanced AI development until a federal regulator sets safety rules.
What is recursive self-improvement?
It is an AI that improves its own design, then uses the improved version to improve itself again. Each round could be faster than the last. Researchers worry people would lose the ability to keep up or step in.
How many Americans support pausing advanced AI?
A Data for Progress poll of over 1,300 likely voters found 68% support and 25% opposition. Backing was 72% among Democrats, 70% among independents and 63% among Republicans.
Has any AI company agreed to slow down?
Anthropic has taken one step. After CEO Dario Amodei’s September 2026 essay, the company said it would give independent outside reviewers staff-level access to check its safety work, without waiting for a law.
