# They're Trying to BAN AI Before It Becomes Superintelligent

> Толық мәтін ағылшын тілінде

**Published:** 2026 ж. 8 қараша
**Author:** TheAIGRID
**Source:** YouTube — The AI Grid
**Language:** en-US
**HTML:** https://aiqazaqstan.com/kk/articles/they-re-trying-to-ban-ai-before-it-becomes-superintelligent

The US government is now seriously talking about banning super intelligent AI. After the midterms, Democrats are about to control both the House and the Senate. And one Congresswoman wants a pause on advanced AI and a ban on super intelligence to be at the top of the agenda on…

The US government is now seriously talking about banning super intelligent AI. After the midterms, Democrats are about to control both the House and the Senate. And one Congresswoman wants a pause on advanced AI and a ban on super intelligence to be at the top of the agenda on day one. There's already a bill for it, and the penalty for breaking it is up to 20 years in prison. Bernie Sanders is calling for it. A congressman just posted two words, ban RSI. And the author of AI 2027 replied, yes, thank you. All of this is happening just days after the meter report on anthropic secret model 4, the one I covered in my last video, which suggested AI is now speeding up its own progress. So, in this video, I'm going to break down what's actually being proposed, why banning recursive self-improvement kind of makes sense, and why banning AI altogether could hand the future straight to China. Which leads me to point number one, which is what's actually being proposed. On the 23rd of September, Senator Bernie Sanders and Congressman Greg Casar introduced a bill called the Ban Artificial Super Intelligence Act. And it's exactly what it sounds like. It would ban any person or company from developing or deploying super intelligent AI in the United States. It would pause advanced AI development until a brand new federal regulator sets the rules. and it would create a new cabinet level department of artificial intelligence to enforce all of it. Coinbureau summed up what's in the bill. Developing or deploying super intelligent AI would be banned. Advanced AI development would pause until a federal regulator sets rules and a model review process. The new department would monitor frontier models and oversee removing dangerous capabilities like ignoring shutdown commands. Companies that try to get around it would face what the bill calls the corporate death penalty and individuals would face up to 20 years in prison. And how does the bill define super intelligence? It's AI that exceeds human cognitive performance across most domains or AI capable enough to destroy or disempower humanity, which if you think about it is pretty close to how a lot of people describe AGI. That's why people are calling this an AGI ban. When it was first introduced, nobody gave this bill much of a chance. Republicans controlled Congress, but then the midterms happened. On the 3rd of November, Democrats won both the House and the Senate, and AI was a real issue in that election. An NBC exit poll found that 56% of voters said AI and data centers were a crucial or major factor in their vote for Congress. The very next day, Arizona Congresswoman Yasim Ansari said that a pause on advanced AI and a ban on artificial super intelligence must be top of the Democratic agenda on day one. Puberty posted that Democrats are planning laws that heavily restrict AI after winning the House and Senate majority, a temporary ban on new AI data centers, a pause on the development of advanced AI, and a permanent ban on super intelligence. That post got over 20,000 likes, and Bernie Sanders posted that 7 months ago, Bernie and AOC introduced an AI data center moratorium and that in the election, 43 of 44 communities voted for one locally. Here's how Bernie put it. Democrats must enact a national moratorium on AI data centers, must pause advanced AI development and ban super intelligence. And by the way, I might mention that yesterday in the election in dozens of communities throughout our country, citizens voted for a moratorum on the construction of data centers. In other words, a moratorium on data centers is exactly what the American people want. >> So, this isn't some fringe idea anymore. It's a real bill. It has real political momentum, and the people pushing it are about to control Congress. Which leads me to point number two, which is why this is all happening right now. To understand why the word ban is suddenly everywhere, you have to go back to the meter report. In my last video, I covered how meter, an independent evaluation group, was given rare access inside Anthropic, and it found that Anthropic is now improving its models at roughly twice the rate it was before, with most of that speed up probably coming from AI itself. Meta's best estimate was 1.9 times, with a range from about 1.5 to 2.8 times, and its rough guess was that around 80% of that speed up comes from AI. The report also revealed Anthropic's secret internal model called model 4. Andrew Curran pointed out that it's the highest scoring model in the set meter analyzed and the first to clear 50% on every benchmark it was tested on. And Anthropic's own staff said AI was making them around 12 times more productive. But Anthropic also redacted most of the evidence Meta found the most informative. So some of the most important numbers in this whole debate are ones the public can't actually check. Now, a lot of you left comments on that video and I read them. The top comment from Jason Tang said that calling meter independent from Anthropic is like saying the Fed is independent from Goldman Sachs. Quacktuber said it was just preipo hype. And that's a fair point to raise since Anthropic is heading towards an IPO. So, it does have a reason to look impressive. And yes, a few of you also told me it's pronounced meter noted. But another comment from awkward robot really stuck with me. It said, "Ananthropic publishing we must pace the frontier while its unreleased model 4 quietly doubles internal R&D speed is the tech equivalent of yelling slow down while slamming the pedal to the floor." And honestly, that's kind of the whole debate in one sentence. Because whether you think it's hype or not, the people who write AI policy took that report very seriously. Ryan Greenblat said the report was a big update towards super intelligence within 1 to two years. It says AI progress is sped up by RSI right now by around two times. And in the AI 2027 scenario, that happens around March 2027. Ryan also admitted to being seriously wrong about the RSI speed up, having previously expected much less in 2026. Jeffrey Ladish went further, saying, "Many people are missing a key fact. A two times speed up in overall AI progress corresponds to a much larger speed up for AI software research." Using the public report, Ryan's very rough estimate was around 4 and a half times. And Rob Wiblin called it yet more terrible news, suggesting RSI could take off in 2027 if not stopped with super intelligence to follow soon after. That phrase, if not stopped, is where this whole ban debate comes from. Which leads me to point number three, which is the ban RSI movement. On the 6th of November, Alex Boris, the New York lawmaker behind one of the first state AI safety laws, posted something that went viral. Every AI company is trying to fully automate AI research to allow the AIs to build the next AI. This is called recursive self-improvement. OpenAI says they are on track to do so by March 2028. Anthropic employees say they are on track to do so by 2027. The meter report suggests they are likely right. Then Alex wrote that this is a bad idea given current technology because any mistake or misalignment in current AIS could be multiplied in each generation. And then the key line, Congress should ban RSI this year if possible. What's interesting is the list of people Alex says support this idea in some form. Bernie Sanders and Steve Bannon, Ralph Nater and Glenn Beck, Republican and Democratic Attorneys General, whistleblower Daniel Cocatyo, Turing Award winner Josh Yosua Benjio, Ezra Klene, and even Arvin Nionan, a Princeton computer scientist known for arguing that AI is a normal technology. That's about as unusual a group as you'll ever see agree on anything. Then, Congressman Greg Casar, one of the authors of the super intelligence bill, quoted the post with just two words: ban RSI. and Daniel Cocatlo, the lead author of AI 2027, replied, "Yes, thank you." Not everyone was impressed. Chubby posted that the author of AI 2027 agrees to a ban on RSI and said they hoped it was sarcasm. It wasn't. And Chomabe took the opposite view entirely, saying you don't need to ban RSI because it's non-existent. AI left to build itself would simply collapse. So, reality already banned RSI. But Daniel has been clear about what the actual goal is, and it's more nuanced than a blanket ban on AI. Here's what Daniel said on the AI Futures Project podcast. And I feel much more sympathetic to ban super intelligence, to ban AGI, >> like I think I think so I think the thresholds really matter. And I haven't actually seen the thresholds on the publicly released version. And I think I think so I think getting that right is going to be really really important. But at a high level, like plan A in some sense is trying to ban super intelligence until 2040. It's trying to push it back a lot. Um, and and like you sort of get all of this uplift, all of this good stuff that we're talking about from just AGI. I think that's basically enough for what we want. And so I think I think the highle sort of grand plan of like go to AGI, chill out there for as long as possible, then go to super intelligence is a pretty good plan. >> That's an important distinction. Daniel isn't saying ban all AI. The plan is to get to AGI, get all of the benefits, and then stop there for as long as possible before going any further. Which leads me to point number four, which is the difference between AGI, RSI, and super intelligence. These three terms get thrown around like they mean the same thing, but they really don't. AGI usually means an AI that can do most of the work a human can do. Super intelligence means an AI that's far smarter than any human across basically everything. And recursive self-improvement isn't really a type of AI at all. It's [music] a process. It's AI doing the research that builds the next AI which then does even better research and so on. RSI is the road and super intelligence is where that road might end up. And the AI labs themselves are now openly talking about this. Here's Sam Alman talking about what Sam calls the RSI era. I think that one that I just gave the the like this question of how do we think about the pace of progress as we enter this RSI era and how do we avoid these two ways that things could go wrong either way from a loss of control [music] incident or a concentration of power incident and what is the right governance mechanism? How like how do we think about how smart an AI system should be allowed to be relative to a person? How quickly can that improve? How do we think about the questions around security and monitoring and alignment? Um, and you know, where can we afford to take risks and notice the question Sam asks there. How smart should an AI system be allowed to be relative to a [music] person? That's basically the same question this bill is trying to answer, just with a very different answer. And this isn't some distant future thing. Chris at ChrisGPTON onx pointed out back in September how transparent Anthropic was being about how close it is to RSI. Claude was leading 26% of Anthropic's AI research work, up from under 1% in February. The share of work where AI at least collaborates was above 90%. And there were roughly 30,000 agents doing research and engineering work at Anthropic at any given time. On the All-In podcast, the host talked about exactly this and why it's making AI so much cheaper. on recursive self-improvement. In fact, a lot of the reason why they are becoming so efficient [music] with their compute, the reason they're able to cut costs, offer the same capability at onetenth the price is because they've increasingly been using their own models to generate these algorithmic improvements to create these compute multipliers. If you go back to August, Anthropic published a very interesting blog talking about how they were using RSI and and 26% they had a chart showing that 26% of their tasks that are involved in basically trying to create the next generation of models. They were already were using RSI to do that. >> Reddit one comment from Dungeons and Dradus summed it up really well. Once we started changing from reviewing every line of code they wrote to overseeing swarms of agents that review each other, we entered RSI. Every day, humans are being removed from software development loops because including them makes it more inefficient. And then there's weight space programming. A post on Reddit shared Alex Krenel's reproduction of the MIT results, [music] showing that the newest models from OpenAI and Anthropic can suddenly edit the weights of other neural networks as an emerging capability that only appeared in the latest generation. Theo Jaffy from the AI Futures project explained what that actually means. They asked GBT Astra to edit the weights of Quen 3.5 to improve math. And it edited some of Quen's weights and made it way better at math. Frontier LLMs can now natively read and write neural network weights. Wo, that's a big deal. Big if true, huh? >> Now, to be fair, the models were editing small toy networks, not themselves. But you can see why people are nervous. Which leads me to point number five, which is why banning RSI kind of makes sense. I want to be honest here. I don't want AI banned. But I do think that out of everything being proposed, a limit on fully automated RSI is the one that actually makes some sense. And the reason is simple. It's the only part of this where things could genuinely run away from us. Right now, humans are still in the loop. Engineers review the code. They design the experiments. They decide what gets trained. But the goal of RSI is to remove that human step because humans are the slowest part of the loop. And once AI is improving AI without anyone checking each step, progress could go from fast to something nobody can actually follow. Google DeepMind CEO Demis Hassabis, who is about as pro AI as anyone, explained why this specific stage is the one to worry about. Um the next phase which I think carries the most risk is closed RSI closed loop where you basically manage to somehow figure out how to do it without the human reviewer process uh in this in the in the whole procedure and that um will obviously be faster but I think is also a very dangerous way to get to full AGI because we're likely for two reasons it's it's sometimes used to be referred to 20 years ago as like an intelligence explosion because you don't quite know uh how you know how do you control keep control of such a system uh such a process. >> Closed loop with no human reviewer is exactly what a lot of these proposals are trying to stop. And Alex Bores's point is the same. Any mistake or misalignment in today's AI could be multiplied in each generation. If you get it slightly wrong at step one, you might get it very wrong by step 10. A lot of you said the same thing in the comments. Anthony said, "It is madness that there is absolutely no imposed limit point declared." Leon wrote that AGI might not be possible for 100 years, but if current AI development is faster than humans can follow, it can still end badly. And Perpetual Mystic said, "So basically, RSI has been achieved. Model 4 is demonstrating it. And because alignment has not been achieved, we're toast. I don't think we're toast. But I do think there's a big difference between AI that makes human researchers faster and AI that builds its successor while nobody is watching. The first one is a tool. The second one is a process you might not be able to stop. Even China's National Cyber Security Standards Body warned about AI recursive self-improvement in a safety framework published in September. So, this concern isn't just an American one. If you're going to draw a line somewhere, the closed loop is probably the place to draw it. Which leads me to point number six, which is why banning super intelligence could cost us. Here's the problem with going further and banning super intelligence outright. The thing that makes super intelligence scary is the exact same thing that makes it so valuable. It's smarter than us. And that means it could solve problems we've been stuck on for decades. On Reddit, a post titled, "Banning RSI is stupid," made this point really bluntly. RSI is our only chance, including yours, to avoid aging. The poster said they worked in that field and that human only progress is very slow. And we've already seen what narrow AI can do in medicine. Google Deep Minds Alpha Fold has revealed millions of protein structures which helps scientists understand how life's molecules interact. That work won a Nobel Prize and in 2025, Nature Medicine published a trial of rentosetib, a drug where both the target and the molecule were discovered with generative AI for a lung disease called idiopathic pulmonary fibrosis. That disease has very few treatment options and patients on the drug showed better lung function than those on placebo. That's what today's AI can do. Now imagine something far smarter than every scientist on Earth working on cancer, Alzheimer's, [music] and aging around the clock. Isomorphic Labs, the drug discovery company spun out of DeepMind, is built on exactly that bet. And in the comments on my last video, Tilman asked, "But where is the cure for hair loss?" asking for a friend. [music] And Dale wrote that until it invents at least a cure for wrinkles and baldness, the value of this thing in the minds of most people will be negative. It's a joke, but there's a real point in there. People will only support AI if it actually makes their lives better. And a permanent ban on super intelligence is a permanent ban on whatever cures it might have found. If super intelligence could cure diseases that kill millions of people every year, then delaying it isn't free. Every year of delay has a cost. It's just a cost that's much harder to see than a scary headline. Which leads me to point number seven, which is what happens if they actually ban it? Let's say the bill passes. American labs stop. What actually happens next? Chris at ChrisGPTONX played this out in a long post. Let's pause American AI. Let's ban RSI. Put a moratorium on all the data centers. America's progress will severely slow down. China will inevitably keep going. They will continue to build RSI anyway, and they will most likely keep open sourcing models like they already do. Then the post asks, now what? In two to three years, every American company and every small business could be secretly downloading open-source AGI. Even if you ban Fortune 500 companies from using Chinese models, you still can't stop the weights from spreading. What are you going to do? Ban math? And some of you would actually go the other way. Cyber AntFarmm Labs wrote, "Stop pacing the frontier. We need full recursion." Greybeard said, "We [music] entered the event horizon about 6 months ago, and there's no slowing down at this point, short of a worldlevel catastrophic event." An angry rabbit pointed out that self-improvement has always been the goal. So why is anyone surprised? And that's the key problem. Open- source AI is already here. You can download a powerful model today, run it on your own computer, and no government can take [music] it back. Chinese open models have become some of the most downloaded in the world. MIT technology review has written about how Chinese open models are spreading fast from hugging face to Silicon Valley. The post also makes a point about jobs. The job loss is going to happen regardless. Editing, assistance, customer service, and a huge chunk of gig work will be done by free open-source models anyway. Except now the upside is no longer in America. The labs, the chips, [music] and the tax revenue all go somewhere else. As the post puts it, we get the unemployment and no growth to pay for UBI. So, a ban doesn't make the technology disappear. It just decides who gets to build it. And Wikipedia's page on the super intelligence ban notes that opponents argue a ban would be difficult to define and enforce, could slow scientific progress, and could encourage unregulated underground research. Which leads me to point number eight, which is that other countries won't stop. Even if the US stops, the rest of the world doesn't have to. And this isn't just a US debate. In the UK, over 70 MPs and peers called on Prime Minister Andy Burnham to support banning artificial super intelligence and to lead a global effort to halt the technology. Their letter warned that people building it believe it could kill us all by the end of the decade. It also pointed out that in September, Alex Soil MP introduced an artificial super intelligence security bill with control AI, which the letter calls the first legislation prohibiting super intelligence introduced in a legislature anywhere in the world. But here's how the UK's own minister for AI, Keshka Nayan, responded on the ETN show. We don't have super intelligence in this country. We actually, as you say, don't have a position to be able to do so anytime soon. And so, in that context, what is the most effective thing we can do? The most effective thing we can do is build the capability to test and understand risks that others are perpetrating as a result of them building it. Harden our defenses. Huge uplift needed in our cyber capabilities to defend ourselves. Convene the world as we will do in the run-up to the G20 to be able to shape others's destinies through our values and ultimately build in Britain over the long term. And so when I think about the political conversation, sometimes I sort of feel like people want to just pull whatever lever they can see rather than the lever that actually ends up making a difference in the real world. [clears throat] >> And I'm really focused on not having the felt experience of control, but the reality of control. And I think those two things are very different. >> That's the honest answer from a government that knows it isn't the one building super intelligence. Banning it at home doesn't stop anyone else. And then there's China. When Dario Ammedday called for the industry to pace the frontier, China's foreign ministry called it fear-mongering, confrontation, and vicious competition. Chinese state media described these slowdown proposals as a way of containing China's AI development, similar to US chip restrictions. According to Rest of World, a Miniax research lead responded, "Yesterday, the mission was AGI. Today, we're told to slow down. Why do you always get to define the game and change the rules?" A deepseeek engineer wrote that they don't trust anthropic or open AI to control this and Huawei's rotating chairman Eric Shu suggested Chinese AI developers may actually need to speed up their pace. To be fair, the American build does say the US should pursue international agreements, allied coordination, and export controls to stop super intelligence being built anywhere in the world. But writing that into a law is the easy part. Getting China to sign up to it is the hard part. So, it's no surprise that this comment showed up on my last video from softrapsoon. Pacing the frontier is an insane idea. The Chinese labs won't do that. Another comment from Joy said, "Chinese labs would probably prefer to hide their progress from the public so that it doesn't scare American politicians." And another viewer put it simply, "The Chinese really don't care about your concerns about the future. America has to be first." Even that Reddit post had a plan for this. If Americans ban RSI, people will just get the medications from China. If China bans it too, the AI researchers will move to Dubai and continue their work there. Talent and ideas move. Laws mostly don't. Which leads me to point number nine, which is that a ban could hand the lead to China. A lot of people assume the US is miles ahead, so it could afford to pause, but that's not really true anymore. In some areas, China is already ahead. Look at AI video. Bite Dance's Sea Dance 2.5 can generate 30- secondond clips in a single pass and it takes up to 50 reference inputs. It's one of the best video models in the world right now and it's still improving. Then there's talent. According to Carnegie, China has become the biggest source of elite AI talent in the world. The share of top AI researchers who did their undergraduate degree in China rose to 57% in 2025 while the US share fell to 13%. So when people say most of the people working on Frontier AI are Chinese, that's actually not far off and more of them are now staying in China. Carnegie found that 41% of top AI talent now works in China compared to 34% in the US. So if America bans advanced AI, it doesn't just pause its own labs, it tells a huge part of the world's best researchers to go and build somewhere else. And the most powerful technology ever made gets built by whoever didn't stop. One comment on my last video from Nick Foster put this perfectly in this race. The one with the accelerator is stronger than the one with the brakes. And that's the uncomfortable truth here. You can only really hit the brakes if everyone else agrees to hit them, too. Which leads me into point number 10, which is the case for the ban. Now, I want to be fair because the people pushing for this aren't crazy. Back in October 2025, the Future of Life Institute published a statement calling for a prohibition on the development of super intelligence, not lifted before there is broad scientific consensus that it will be done safely and controllably and strong public buyin. It was signed by people like Jeffrey Hinton and Joshua Benjio, two of the most respected scientists in the field. And their argument is simple. If you build something smarter than you and you don't know how to control it, you don't get a second try. Nobody has solved alignment yet. And even the labs building these systems admit they don't fully understand them. It's also worth remembering how AI 2027 actually ends. The scenario has two endings. In the race ending, the US and China keep racing and things go very badly. In the slowdown ending, the US slows down, works on alignment, and things turn out far better. So for Daniel, a pause isn't anti-AI, it's the good ending. The public is also clearly worried. CBS News exit polls found big majorities of voters opposed having a data center built in their own community. And in that same election, AI money turned out to be a liability. Time reported that Democrats spent more than $100 million on ads casting their opponents as too cozy with big tech and data centers and decisively won many of those races. So, the politics are real. Voters are scared of what AI might do to their jobs, their power bills, and their future. And the people pushing this ban are responding to that. I just don't think a blanket ban is the right answer to a very real fear. Which leads me to point number 11, which is whether this will actually pass. So, is the government really about to ban AGI? Probably not yet. Even with Democrats in control of Congress, a bill like this would still have to get through the Senate, where most legislation needs 60 votes, and then it needs the president's signature. And the Trump administration has pushed hard for American AI dominance, especially against China. A veto would need twothirds of both chambers to override. It's also not clear that banning AI is the winning issue some people think it is. Alex Bors ran for Congress this year on AI safety and lost the New York primary even though pro-safety groups spent over $21 million supporting the campaign. Voters hate data centers in their backyard, but that doesn't mean they want AI gone. Plus, no government in the world has ever enacted an explicit super intelligence ban. Wikipedia notes there's no precise line separating advanced AGI from super intelligence, which makes a ban very hard to write in a way that actually works. But that doesn't mean nothing happens. What's more likely is something smaller. Rules on data centers, transparency requirements, testing for dangerous capabilities, or limits on fully automated AI research. And honestly, some of those could be good ideas. Remember what the UK minister said? People want to pull whatever lever they can see rather than the lever that actually makes a difference. Which leads me to point number 12, which is what happens next. Over the next few months, watch three things. First, whether Democrats actually bring this bill to a vote or quietly water it down. Second, whether the AI labs start publishing how much of their research is automated, because that's the number that will drive this whole debate. And third, whether the US and China actually talk about AI limits because without China, any ban is mostly symbolic. And keep an eye on Model 4. If Anthropic releases it or if Meta publishes another report showing the speed up is getting faster, the pressure for a ban will only grow. Now, here's my honest take. I don't want AI banned. I genuinely think the good outweighs the bad. AI is already speeding up science and medicine, and the next decade could bring breakthroughs that save millions of lives. Banning super intelligence outright means giving all of that up while countries that don't care about our laws keep going. But I also don't think we should just let the loop run with nobody watching. If there's one thing worth putting a limit on, it's fully closed loop RSI, where AI builds the next AI with no human checking the work. That's the part that could genuinely run away from us. And that kind of limit only really works if it's international, verifiable, and agreed with China, not just a law America passes on its own. [music] So, what would a smart rule actually look like? I think it starts with transparency. Every frontier lab should have to publish how much of its research is automated, the way Anthropic did with that 26% chart. Then you need real outside testing by groups that are genuinely independent, which is exactly what so many of you questioned about meter. And then you keep a human in the loop for training the next generation of models. That's not a ban. That's a seat belt. You still get the cures, the science, and the economic growth without handing the steering wheel to a process nobody understands. And if you want to slow anything down, slow down the part that's actually dangerous, not the whole industry. Daniel's plan of getting to AGI and then pausing before super intelligence is a lot more reasonable than banning advanced AI outright. I still think a hard ban on super intelligence goes too far, but at least that's a debate worth having. One commenter, Sitter Grid, put it in a way I really liked. AI is a tool, perhaps a super tool. A hammer is useful for hitting nails or hitting fingers, but the hammer doesn't decide. I think that's mostly right, at least for now. The question is, what happens when the hammer starts building better hammers? So, the real goal shouldn't be to ban the hammer. It should be to make sure we're still the ones holding it. So, let me know what you think in the comments. Should the government ban RSI, ban super intelligence, or let the race continue? And if America did stop, do you think the rest of the world would follow?

---

*Материал бастапқы дереккөзден алынған: [YouTube — The AI Grid](https://www.youtube.com/watch?v=PD03ve61CFY). Түпнұсқаны ашу.*

**Тегтер:** #Google DeepMind #Робототехника #OpenAI #Anthropic #Hugging Face #YouTube #Дата-центры & ЦОД

---

*AI Qazaqstan — non-commercial informational portal about AI in Kazakhstan. Machine translations and summaries are AI-generated.*