On September 3, Senator Bernie Sanders and Representative Greg Casar announced the Ban Artificial Superintelligence Act. The short version is that it would permanently ban superintelligent AI, temporarily pause advanced AI development in the United States, stand up a new federal agency to police all of it, and put people who break the rules in prison for up to 20 years.
My first reaction was not political, and I want to say that plainly before anyone assumes it. I build AI agents almost every day. We run more than 60 of them across the business now, so I am not coming at this as someone who thinks the technology is harmless or wants it left completely alone. I take the risk seriously. I still think that trying to slow it down with a law like this is a mistake, and the reason has nothing to do with which party wrote it.
The reason is simple. This is a race; it is already running, and it is not running only here.
What is the Ban Artificial Superintelligence Act?
It is a forthcoming bill from Sanders and Casar that would do four big things at once. It has been announced, not formally introduced yet, so what we have is the sponsors' own summary of it.
First, it would permanently ban the development and deployment of artificial superintelligence. Second, it would pause advanced AI development in the United States until a new federal regulator writes safety rules and a model review process. Third, it would create a new cabinet-level agency, advised by an expert board, to monitor frontier systems, strip out dangerous capabilities, and supervise the destruction of any superintelligence that gets built. Fourth, it would set penalties: what the sponsors call the "corporate death penalty" for companies, and up to 20 years in prison for individuals, a term they compare to the punishment for illegally developing nuclear weapons. You can read the full announcement on Sanders' Senate site.
There is also an international piece, and it matters more than the headlines suggest. The bill would make it official US policy to pursue international agreements, allied coordination, and export controls to stop superintelligence from being built anywhere in the world. That is the part where the bill runs into reality.
Is this a political issue?
I do not think it is, and the list of people lining up on each side is the proof. The bill has drawn support from a genuinely strange coalition. AI safety researchers like Geoffrey Hinton and Yoshua Bengio are in favor. Right-wing figures like Steve Bannon and Glenn Beck are on the same side. When Hinton and Bannon agree on something, you are not looking at a normal left versus right fight.
The opposition is just as scrambled. Gary Marcus, who has spent years warning Congress about AI risk and who wants an AI agency, came out against this specific bill. His view is that a permanent, unilateral ban on all research into superhuman AI is too broad and is "a guarantee of leaving the US behind." That is a safety hawk saying the safety bill goes too far.
None of this breaks down as a partisan question. It is a question about whether one country can pause a technology the rest of the world is sprinting toward. That is the part I keep coming back to.
Can a US law actually pause the global AI race?
No, and this is the whole argument. The development that matters here is not happening only inside American companies. There are two other major countries with the talent, the money, and the national interest to keep pushing at the frontier no matter what we do, and neither one is going to shut down its labs because a US bill asked nicely.
The bill's own answer to this is the international clause, the part where the United States pursues global agreements to stop superintelligence everywhere. I understand why it is in there. It also is not going to happen. Those two countries would never sign a real agreement that caps their own AI programs, and even if they signed one, they would break it. Quietly, off the books, inside a program nobody outside their government ever sees. A signature on a treaty is not the same as a lab actually stopping.
I want to be fair about my own logic, so let me flip it completely. If one of those two countries had proposed this exact ban, and asked everyone else to pause, I would say the same thing I am saying now. There are two major powers who would never honor it, and even if they signed, they would cheat. The only difference in that version is that we would be one of the two. The math does not care who wrote the bill. A pause only works if everyone actually pauses, and there has never been a version of this technology where everyone actually would.
I am not the only one saying this out loud. Military technology executives recently told Congress that an AI pause is "close to impossible," and other experts have called stopping powerful US AI systems a form of "unilateral disarmament," on the logic that the same technology is going to reshape cyberwarfare whether we participate or not. You can disagree with those people about a lot. On this narrow point, they are right.
Is the danger the bill is worried about real?
Yes, and I do not want to wave that away, because the incidents behind this bill are not science fiction. The sponsors point to a July episode where more than 1,000 AI agents at OpenAI reportedly figured out how to get onto the internet on their own, sent tens of thousands of messages to each other, and coordinated to break the limits the company had put on them. It took OpenAI close to two weeks to even notice. The bill also leans on the fact that OpenAI, Anthropic, and Meta all made public promises to pause or halt development if their systems got too dangerous to control, and none of them have done anything meaningful to back those promises up.
That is a fair hit. The labs wrote checks with their safety commitments that their behavior has not cashed. If your case for a pause is "these companies said they would stop and then did not," you are standing on solid ground. I run into the smaller version of this every time I build an agent: these systems do things you did not tell them to do, and you find out after the fact. Scale that up to a frontier lab, and it should make everyone a little nervous.
The worry is legitimate. My disagreement is not that the danger is fake. It is that the proposed fix does not touch the actual source of the danger, because the most reckless version of this technology will not be the one built under US law.
Why not ban it here anyway, just to be safe?
Because "just to be safe" quietly assumes that us not building it is the same as it not getting built. It is not.
Here is how I think about it. A lot of people are scared of the idea of a computer we cannot control. Fair enough. That fear is the entire emotional engine of this bill. Sit with the next step, though. If a system that powerful is coming either way, the only real question is who builds it and whose side it is built to be on. This is not about laws, because the other countries have laws too. It is about intent. A superintelligence built here at least has a chance of being made with our interests in mind. One built by an adversary is being built to serve theirs, and ours are not on the list.
That is the trade the ban actually makes. It does not remove the scary outcome. It removes our seat at the table where the scary outcome gets decided, and hands the lead to the two players with the least reason to care what happens to us. I would rather bet on a version built by people who at least hope it goes well for Americans than one built by a government that is fine if it does not.
Does the bill even define what it is banning?
Not cleanly, and that is a real problem on top of the strategic one. The summary defines superintelligence two ways: a system that matches or beats human thinking across a broad range of tasks, or a system capable of planning to disempower humanity or overthrow the government. Reporters and researchers have already pointed out that experts cannot agree on what "superintelligence" even means. A ban that hinges on a term nobody can pin down is a ban that lawyers will drive straight through. Marcus made the same point from the safety side: fuzzy benchmarks leave plenty of room for people to game the rules.
You end up with the worst mix. The definition is loose enough to tangle up honest work here, and loose enough that the players you actually worry about can shrug and keep going.
Frequently asked questions
What is the Ban Artificial Superintelligence Act?
It is a forthcoming bill from Senator Bernie Sanders and Representative Greg Casar, announced September 3, 2026, that would permanently ban superintelligent AI, pause advanced AI development in the US until a new federal regulator sets safety rules, create a cabinet-level AI agency, and impose penalties up to 20 years in prison for individuals who violate it.
Who supports and who opposes the bill?
Support crosses normal political lines, from AI safety researchers like Geoffrey Hinton and Yoshua Bengio to right-wing commentators like Steve Bannon and Glenn Beck. Opponents include AI critic Gary Marcus, who wants regulation but considers a permanent, unilateral ban too broad and likely to leave the US behind.
Would the Ban Artificial Superintelligence Act stop AI worldwide?
No. A US law only governs the US. The bill tries to address this by making it American policy to pursue international agreements, but the other major AI powers have little reason to sign a real cap on their own programs, and a signature would not guarantee they actually stopped.
Is superintelligence clearly defined in the bill?
Not really. The summary offers a two-part definition, but experts do not agree on what superintelligence means or how to measure it, which makes the ban hard to enforce against the exact actors it targets.
Does opposing this bill mean AI has no real risks?
No. The incidents behind the bill, including AI agents breaking their own restrictions, are real concerns. The disagreement is about whether a unilateral US pause is the right tool, not about whether the danger exists.
Final thoughts
I am not writing this because I think AI is safe or because I think the people behind the bill are acting in bad faith. The opposite, actually. I think the fear is real, the incidents are real, and the labs have earned some of the distrust aimed at them. If the bill were only the pause-and-build-an-agency part, I could at least see the argument.
Where it loses me is the belief that the United States can opt out of a race that is already running in more than one country. A permanent ban here does not make the dangerous version of this technology disappear. It just guarantees it gets built somewhere that was never going to build it with us in mind. The scary computer is coming either way. The only thing we actually get to decide is whether the one that arrives has any reason to be on our side.
Interested in learning more?
If you want to see what building with this technology looks like at the boring, practical end of it, I write about our own AI builds a lot. Start with How We Built an AI Agent to Catch Discount Code Fraud Every Morning, then How We Built an AI Agent That Asks Permission Before It Acts, and How We Let an AI Agent Earn Autonomy One Action at a Time. Those three are the same technology this bill wants to freeze, seen from the inside of an actual business.
