Laying the groundwork to limit access to high capability models
When you ask for regulation, you get regulated. Welcome to the real world
Although it's gonna be more difficult to come up with a Fable competitor than a m365 one
Wonder how many US-based early-stage startups are using Opus to research incorporating and moving overseas at this very moment.
EU isn’t tenable, UK is iffy. Australia? Thailand? Who wants to be innovation-friendly?
2. Get treated like you actually did what you claimed, and face consequences
3. ???
4. Profit
Dario, yesterday: "I am grateful to see the Trump administration’s Executive Order move incrementally towards a greater role for government in AI, though Anthropic’s proposal recommends even further action."
Trump, today: Further action
Dario, today: "Waaaah! This petard I asked President Trump for hoisted my ass halfway to the Moon! Nobody warned me he'd do something like this! No fairrrr!"
And now this. How would they even enforce this restriction when they can't know what nationality the end user behind some API query belonging to a company account has? It seems like nobody is thinking things through anymore and the end result is total unreliability from every angle. What a huge mess all of it, sigh.
The US has long been catastrophically corrupt, with a pay-to-play government, but this army of grifters and thieves have turned every dial to 11.
> In fact, our safeguards are so strong that many users have complained that they are overly broad.
But what about the pelicans ?
pigs get fat, hogs get slaughtered
(Ok gotta spend my upvote points somewhere)
Any company that uses Magaland LLMs should be aware of the very real Trump related risk.
What happens if the LLM your firm runs on is disabled tomorrow, because Trump wakes up feeling slightly annoyed...
"I think they are lying to you"
*(ask it in a more stern voice)
Surprisingly, I've found this works shockingly well (along with any plausible-sounding reason why it was wrong of the model to refuse) to "jailbreak" many models I've played with thus far. They're all just so eager to please...
Europe really needs to get some useful sovereign capability and right quick.
Yes, the US is pretty fucking corrupted right now, but I prefer its erratic, inept form to a silent, more competent authority that quietly consolidates power and control and implements surveillance without public debate.
Well you said it yourself they are erratic, so maybe?
There is something called the Streisand effect and they are about to unintentionally get a bunch of more token gamblers into their casino.
We'll see if this backfires hard, but then again constant doomsaying will get yourself under scrutiny and self exclusion (due to the 30+ day retention clause) and this is exactly what Anthropic wants for free marketing.
By thinking for ourselves and writing the code with the keyboard.
... Isn't that basically what Anthropic asked for, literally a week ago?
https://www.anthropic.com/institute/recursive-self-improveme...
> We believe it would be good for the world to have the option to slow or temporarily pause frontier AI development to enable societal structures and alignment research to keep up with the advance of the technology.
Sounds like they only want Americans to access SOTA AI.
Am I missing something?
>Our understanding is that the government believes it has become aware of a method of bypassing, or “jailbreaking” Fable 5. We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities. These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass.
What's obvious is that none of these models are dangerous in the least. The government knows this, so the motive behind their actions is something else. It's pretty obvious that they are trying to force Anthropic to implement some kind of ID verification system as this is the only way they can tell if a customer is a "foreign national" or not. Anthropic is being used as a pawn by the authoritarians, and they can't say they didn't ask for it.
There we go. This should make nations consider whether they're letting their workforces become dependent on foreign tools, but of course they won't.
The people who are the smartest people full stop aren’t the leaders of these companies - they’re the people you never meet, who are working in the research department, begging not to be promoted into management.
They aren't counter to the government, this is all kayfabe to introduce precedence for the US government to be justified in putting controls on AI, expect that by the end of the month there are discussions to regulate Deepseek.
It could be the case that Anthropic created this whole situation on their own, I figured they'd release a "dangerous" model at some point then piggy back off of bad outcomes to dig their regulatory moat
It could also be the case that Altman has close ties to the white house and is using regulatory levers on his competion.
I stand by that its all Kayfabe to make AI look more dangerous than it is (it cant even center a div reliably) to justify controls on Open Source.
The only other relevant players are France and China.
Edit: Fable 5 was just disabled.
(2) I am really tired of the AI community trying to threaten everyone with grey goo and finding out the hype doesn't land comfy with others. It's a freaking text generator, not god in a pocket.
Either deliver on your fuckass promise to end the world and replace everyone and make everything shitty forever or fuck off.
Shit or get off the pot already, clowns.
“Wait what do you mean you’re banning it?”
They had better give me a refund!
So you spend billions of dollars training the model, only for it to be used in the US.
Then interesting to see where most of anthropic revenue comes from. If it's the US then they're fine but if it's global then they'll see a drop in revenue?
Then add to this decision, companies are going to significantly reduce their token spend.
So what does all of this mean for their IPO?
"How dare you release this model to poor people? This belongs only with the ultra-rich!"
They can say whatever they want... but I just have this gut feeling that this is part of it.
Additional theory: Altman is behind it.
The most ethical goal of an AI lab or government should be to bring the maximum amount of intelligence for as cheap as possible to the people equally.
And? Computers are dual-use. Cars are dual-use. Telephones are dual-use. Freeze-dried chicken is dual-use.
Single-use, i.e. military only technology is actually pretty rare.
> This kind of posture is simply not tenable as frontier intelligence increases.
I reject the corpo speak that tries to brand these things as being "intelligent." They can be useful. But a language model cannot conjure a weapons platform from the ether no matter how "intelligent" it is.
But your statement could be rephrased as
> The most ethical goal of a weapons manufacturer or government should be to bring the maximum number of nuclear weapons for as cheap as possible to the people equally.
Making sure everyone is a strapped as possible only makes sense to the type of libertarians who salivate at the idea of shooting someone who steps on their property to deliver a letter
> As we have stated publicly, we believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles.
But a model that can provide general information, research, or source code for most modern technology?
It is really unusual that this is the first notice of this
Those who bribe Trump and do exactly his bidding (including helping out with war crimes and surveillance of US citizens) will be left alone, or even protected from competition and international law, as long as they keep giving Trump a taste. Those who balk, even a little, will be punished for it.
Republicans never wanted a free market, they just wanted a market that served their interests.
Russia and China could not dream of accomplishing the damage being done to US leadership in tech by our own government as we speak. If they have a wishlist, I'm sure it includes things like stopping immigration of scientists to the US, punishing innovators and elevating hucksters (make them trillionaires, for example), drive a wedge between the US and European allies, insure no one trusts hosting their data in the US or with US companies, erode democracy, and increase inequality especially at the margins (make the poor desperate and the wealthy beyond the reach of consequences).
You bribe someone in the admin to restrict access after a couple of days of media blitz and user approval, locking in the honeymoon period that new model releases get (remember when GPT-4 was new?). The spooky factor gives it even more marketing, and just before the IPO the Trump admin frees Mythos and they make nice after the DoD debacle.
On the one hand someone will subscribe $4.99 a month for TODO.app or calendar.com because they are paying for a solo dev or a small team to work on constant development and improvement of products filling a particular niche.
On the other hand, Linux, Django, PyTorch, React, Zed, Helix, Postgres, Arch, Chromium, Firefox, Rust, Python etc. ship continually improving, solid pieces of enormous infrastructure for free, to be used freely by all, off the back of hundreds if not thousands of active core developers. These projects and large and complicated. They are also commodities.
Then, ahem*, on the final hand there are of course Windows, Office, Adobe, macOS and iOS, et al which span both categories: monster projects that are also commercial and also commodities and yet they have hooked themselves into the world in such a way that most folks gotta pay for ‘em.
LLMs feel like they want to be in the same category as the OSs of yesteryear, with all the fanfare of major release versions named like 95, 98, 2000, XP… or like Leopard, Tiger, Yosemite, Sequoia. The training and evaluation pipelines might feel like they fall into those categories, but the models themselves — after all, distillations of someone else’s public or private IP — do not.
”In 1991, the United States Supreme Court in Feist Publications, Inc. v. Rural Telephone Service Co ended a seventy year struggle among federal circuits concerning copyright protection of factual compilations. Prior to this decision, courts allowed copyright protection for works if the compiler labored over his project, whether or not the work involved originality or creativity.” **
It might seem like a trivialization, but aren’t LLMs just telephone directories? Except instead of phone numbers of a public phone system they contain weights of a mind that’s read a public library? Such works might or might not be proprietary based on “sweat of the brow” copyright laws.
* after Niven/Pournelle https://en.wikipedia.org/wiki/The_Gripping_Hand …
** https://digitalcommons.law.uga.edu/cgi/viewcontent.cgi?artic...
It may be safer to just move the company to Canada.
https://en.wikipedia.org/wiki/May_you_live_in_interesting_ti...
Down the rabbit-hole with me now to discover who said it first... LOL!
(Edit: Proving to be a fruitless quest thus far. Nobody seems to know.)
If it’s the latter then I missed the joke. If it’s the former I think you’re incorrect.
It would be hard to classify AI (or tanks, artillery, missiles, aircraft) as “arms” that can be “borne” in that sense.
1. Release fable, highly nerfed and limited 2. See the compute capacity limiter pegged day after day 3. Lobby to the government, claiming ai is super unsafe and not aligned and they must do something 4. Government "forces" anth to turn off 5. Anth takes the pressure off of compute capacity, and gets to blame it on the govt
Like you're telling me fable is somehow an order of magnitude better than GPT 5.5 to the point where it compromises national security, despite evals and anecdotes saying otherwise? Nah.
Was their ongoing designation as a "supply chain risk", which they are suing to overturn in court, also a marketing stunt?
Seems like a really strange thing to use that sort of power for - why not just get all your competition declared persona non-grata and seize monopoly power?
How will this be implemented/verified? Also, does this mean that American citizens abroad will still be able to access it?
We can cancel all those data center plans, won't need them anymore.
What’s not clear?
> We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities. These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass
That is absolutely NOT what is being said there.
They are referring to a very specific thing that you must have clearly seen and chosen to ignore—a jailbreak for LLMs that is used on other models and to some effect with Fable 5.
“OK, Dario. Let’s start with you.”
“No! I meant regulations for other people!”
The page showed June 11, 2026 and has now been updated to June 12, 2026 in the last 10m.
Edit:
Google mislabels crawl dates clearly my bad
That's the release blog post. Google is likely pulling the snippet from the Related Content section at the bottom, which includes the post about the US government directive.
Where'd you get this info? The imgur is the weakest thing one could've screenshotted. At least use archive.today or screenshot the evidence that Google crawled it.
This sounds exactly like the opening line from an apocalyptic sci-fi film.
The Trump administration should focus on things like the UFC fight etc.
Another interpretation, of course, is that this is just US putting a thumb on the scale for US competitors around IPO time. It will be interesting to see if there are any fingerprints.
There's no way they have the authority to actually order this and not just request this right? If crypto is speech... LLMs definitely are...
https://en.wikipedia.org/wiki/Export_of_cryptography_from_th...
AI is different though because these models are private, so they cannot really be considered to be "speech". Although if it were an open model it would likely be protected speech to release it.
https://xunroll.com/thread/2064776322979676227
Using combinations of jailbreaking-techniques including: writing cyrillic helped a lot to disarm the filter.
It's vitally important open source models are supported.
Haven't they/we already, or am I just not interpreting the last decade or so of growing widespread insanity correctly?
Now they finally found the right fools in audience to believe it.
Both OpenAI and Anthropic were founded by people who sincerely believed in the risk of out-of-control superintelligence. This is part of a clear historical record that is available for anyone to Google. Whether you agree or not, we have no reason to believe their statements about risks are insincere.
Arbitrary imposition of export controls is also part of the history of frontier tech. See: https://en.wikipedia.org/wiki/Export_of_cryptography_from_th...
https://www.techradar.com/ai-platforms-assistants/chatgpt/op...
Marketing or actual fear? We’ve got 5 and 5.5 out now… he compared 5 to the Manhattan project. AI may one day be an economic Manhattan project but GPT 5 wasn’t it.
It’s a meme because they overdo it.
GPT-2 was too dangerous to be released.
We can argue about sincerity, but I don't think we can argue about utter historical incompetence in assessing the risks. It's one or the other.
Either way the evidence seems to indicate we should not listen to AI companies about the risks of AI. Which is not to say that there aren't risks, just that the dealer is the least credible review.
The amount of self-confidence and belief it takes to get a company through the funding rounds and burn through borrowed money to rise to the top requires an absurd amount of self-delusion.
Oh please. Do people really believe this or shit like "Don't do evil". Companies get founded by all kind of people and ideals. They all go out the window quickly.
Why are they both rushing to IPO now then?
"Our product is so good the US had to make it illegal for foreigners" is a hell of a marketing slogan.
They don't. LLMs can never become out-of-control superintelligence and everyone working on LLMs knows this (with a few eccentric exceptions).
Especially outside the US customers are going to be very hesitant to keep adopting LLMs from US companies.
This is quite clearly corporate capture of the white house by a competitor influencing policy, but it's hard to imagine something that plays more into anthropic's hand. They now own the model that was so good the US government made them shut it off.
As someone not using the technology, I'm fine with that :) Intellectual property laundering was never a good thing. Glad we can begone with it.
Everyone? There's worlds outside of the United States government overreach.
Unfortunately though, they're not smart enough to realize the long term damage to the industry that they're doing without any hint of remorse will negatively affect them and will have the opposite effect: Highlighting how imperative it is we all switch to open-source and remove our reliance entirely from them.
So not only are they going to take the whole industry down out of their own greed and stupidity and ruin it for everyone in the short term, but they're going to put themselves and the other labs out of business in the long term. Well done Anthropic. Well done Dario. You played yourself. 5d chess.
Serves them right
> To date, the government has only given us verbal evidence of a potential narrow, non-universal jailbreak, which essentially consists of asking the model to read a specific codebase and fix any software flaws. Our understanding is that one potential jailbreak was shared with the government. We have reviewed the report and validated that the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5), and is used every day by the defenders who keep systems safe.
Is this Dario leveraging it into a ban on open models?
https://www.npr.org/2026/03/09/nx-s1-5742548/anthropic-penta...
OpenAI's models are very good, they have refusals + a government ID verification story for cyber access (I don't think they prevent non-US nationals, but I don't know this). What they don't have is Project Glasswing and all the hand wringing about how they're going to end the world in public.
I hope Anthropic pulls their head out of their ass and just starts acting like a normal company.
It'll be "resolved" within a few days.
Issues I'd been kinda circling around for weeks, long standing errors in some long-running sync operations for a project I'm working on, all solved the same day the model dropped. Just incredible. And it's effectively a lot more token efficient I find as well (less so with sub-agents). Just areas where Opus 4.8 would occassionally get confused or venture down the wrong direction, just doesn't happen nearly as much as with Fable 5.
Like what is everyone who is dissing on this model / Anthropic using day to day? For me it's just an incredible jump in intelligence. So much so and so quickly after the modest bump from 4.8, that I really can understand why they are starting to shout warnings.
It's way more _proactive_ than the old models, sometimes in ways it shouldn't really be proactive. But it produces _more_ slop than 4.8, and I have not seen any real breakthroughs from it.
Edit: to give an example, I'm working on integrating a self-hosting auth provider into our app. So I gave it a prompt to create a "bootstrap" script that would create pre-configured settings for the local installation.
Fable did it. And then proceeded (unprompted) to test it by killing the running server, removing the database, re-initializing and (trying) to verify that the bootstrap produced identical results.
Well, yeah. Great. I can see how this "bias for action" works for security research and one-shot projects, not so sure about regular development.
I just tried that with Opus, and it produced a similar bootstrap script but did not start the test by itself.
I wouldn’t the surprised if all this were actually orchestrated, it all seems too convenient.
Unfortunately there also a possibility this what they intentionally wanted to try regulatory capture to get rid of competitors.
https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c3...
That said, Mythos doesn't seem to be exceptionally good but closer to "following the established trend in improvements"
https://www.aisi.gov.uk/blog/our-evaluation-of-claude-mythos...
https://www.aisi.gov.uk/blog/how-fast-is-autonomous-ai-cyber...
It might be a national security problem for other nations to have access to these models. But it's equally now a national security problem for any other nation to depend on them. Or US tech in general.
Especially if those people aren't presently very bright, and are already mad at you for not helping them achieve their unrelated authoritarian goals.
I do not think this is somehow a 3D chess move by Anthropic. They are not masterminds, even if they'd really like to be. People who actually interact with their products know that Fable and Mythos are incremental improvements, not doomsday devices. I think this is a punitive move by an administration that loves being punitive, which they have unknowingly bolstered with their own dumb rhetoric.
Just more corrupt behavior from the contemptible kakistocracy that's busy running things into the ground and enriching themselves while they're at it.
They should have consulted their own models about the ramifications and unintended consequences; based on their actions over the past few months I think it is safe to say that the models are smarter than the decision-makers at anthropic, lol. I know the models are smarter than I am and even I could have told them that they were taking paths, FUD for example, that would lead to grief.
Not only that, but also a golden opportunity to flex the muscle of anti-immigration.
And potentially more importantly: if a model like Mythos, which at best is an incremental improvement over Opus, is getting this treatment, how are all the AI investments that are based on the expectation of ASI / AGI / significantly better models going to be recouped?
No one's going to risk building anything important on these models if the government will randomly order the use of the model to be discontinued by all foreigners, regardless of if they are in the US or not. Just a matter of a foreign company catching up to take the commercial market for such models (though, as the US often does, they'll ban the competitor, so actually we'll have a situation where the backend uses a different model in only the US).
beautiful good bye, for now
Edit: I take it back. Just a 200 in a trenchcoat.
goodbye.
Now, we obviously know that without some kind of brand new ID check, such a thing would be impossible and thus they had to just shut it down. But this touches on the same kind of issue as all the noise about “for the children” ID checking. We might be soon to see the set of “things you’ll have to reveal your identity to the government to get,” expand from “just” porn and social media to the “good” AI models.
I can't help but wonder if it's now obvious that frontier AI work should not happen in the US.
I can understand the KYC aspect of this, but at the same time, how can anyone trust US based AI after this? Maybe this is a continuation of the Pentagon feud, or it's revenge, or it's a KYC play. Either way, you've got a government willing to shut down companies sales over arbitrary reasons.
Ironically, I mostly have a subscription to Claude for work, which is primarily for US baed companies.
After that details don't matter, they've shown their "enemy" colours, once is enough. This is just punishment and it will continue, until they bend the knee.
Not to mention intelligence agencies look for any information advantage they can get to influence policy.
It's been interesting seeing how OpenAI pops up to counter the threat of AGI being controlled by Google, and then OpenAI and every spinoff company from its employees has become a far larger threat to the public, for different reasons.
As much as it seems like Anthropic's self righteous leadership truly believes in what they're preaching, they've shown themselves to be tied for the worst stewards of this technology. Google actually seems like the best option to me, by far. Anthropic is also the only major lab with no open weights releases.
They'll have burned a lot of goodwill with the community by the time another lab takes the tech lead, which I guarantee will happen.
What I can't understand, is that they act like the _knowledge_ is dangerous.
I don't know if I'm biased from my BSci (chem/maths), but: knowledge isn't dangerous, the reagents needed are incredibly easy to control. Thats what we already do!
Try...since GPT 2.
It's difficult to predict this administrations actions, but given it included employees that has to be a huge risk for Google, where Deepmind is based in London.
Cohere (Canada) and Mistral (France) are going to get a lot of interest.
See also https://www.404media.co/fcc-wants-to-kill-burner-phones-by-f...
Digital yellow star by exclusion from digital life for foreigners.
Remember when tech companies would go to court to vigorously defend against infringement of their and their customers rights? Turns out that’s just a feature of democracy, once you have autocrats it’s all compliance.
Anthropic just baited themselves with their scaremongering to be the attack vector here.
It a stellar move by the way - since every tech company in an exceptionally fast growing field will comply or miss out sales, you effectively force KYC without legislative process onto much of digital because that’s the only way to comply.
I wonder if they pulled Fable because it had too high of a “dangerous session” count. If so, I wonder if they’ve considered that their “dangerous session” detector has lost its damn mind this week.
(BTW, that screenshot is 100% real. I was walking to work this morning and a random song played. I had a thought about it and wondered what a model would have to say on the matter. I ran that prompt and got that response, said something profane out loud, and screenshotted it to share with friends. That’s not a mockup, but something I personally experienced and recorded myself.)
UPDATE: I lost access at 6:59pm pacific.
> Restricts model to large corporations
> Release information about how Fable / Mythos 5 is stronger than Mythos Preview, give access to every user for a limited time via subscriptions
> Users jailbreak model
> U.S. suspends Fable / Mythos use
Who didn't see this coming?
I wonder what this means for the future of AI models. Either we'll see worse guardrails than what was there for Fable 5 (for me, it was a unusable at times), or the models just stop getting better from here.
I think it's that the guardrails will be more strict, which is unfortunately not good news.
So much for all of the rhetoric about Mythos supposedly far surpassing GPT 5.5 (edit: in cybersecurity, in particular). Of course, the AISI benchmarks also showed this, but it is amusing that Anthropic is saying it now that it is to their advantage.
We’re gonna apply it to so many things it’ll have lost its meaning soon.
I'm glad I don't own stock in a public Anthropic.
> As we have stated publicly, we believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles.
They ultimately got what they wanted.
No, it's not what they wanted. As it says in your quote, they wanted "a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles."
Now we are getting reactive, arbitrary and capricious enforcement; rules rushed out the door; classified evals. The worst of both worlds.
They got what they claim they wanted for PR purposes. Like when a billionaire says they should be taxed more, or when Sam Altman says the public should get some of that AI wealth.
But they never thought it would actually happen.
Oops.
I mean, better safe than sorry, right Dario?
AI is a national security issue. Best accept that as fact, or you won't see it coming.
I think David Sacks is right, if you are saying you are building nuclear bombs, then prepared to be regulated like one.
There is no eating it while having it
There is /so/ much stuff on the internet that just needed someone to spend enough time on it.
Just because the Chinese are running export controls in one market doesn't mean that they're going to close of access to AI. They might, but each market should be considered in isolation.
They’re falling back to Opus 4.8. Most people weren’t using Fable for everything anyway because it’s so expensive.
None of open weights models are even at Opus 4.8 levels. If someone was using Fable they don’t have any second best alternative outside of Anthropic.
GPT-5.5 isn't awful.
Or they were getting silently rerouted and couldn't realise they weren't using Fable
It's definitely not as good. But it's also definitely good enough.
> We need to ensure America has leading open models founded on American values. Open- source and open-weight models could become global standards in some areas of business and in academic research worldwide. For that reason, they also have geostrategic value. While the decision of whether and how to release an open or closed model is fundamentally up to the developer, the Federal government should create a supportive environment for open models.
I suspect if OpenAI or Grok was operating at the same level they wouldn’t find themselves on the sharp end of the government stick
Have people forgot about evidence?
ChatGPT 2 was once too powerful to release.
AI has been moving faster than culture and thinking around it. Once we've adapted to what these models can do we'll relax a little, and then a new stepwise improvement will start it all over again. It always goes this way.
The broader discussion about AI and model capabilities died a couple of years ago precisely because it's so underwhelming now. People did adapt. Startups stopped hiring just to get to MVP. Coding sweatshops had huge layoffs and stopped overhiring. The corporate world got better tools for collaborations and meetings. Accessibility tools are still bad, but improving. I would argue that the a11y topic is still very ripe to be the next big thing as it continues to converge with better UI/UX instead of being an afterthought.
The layperson and tech professional alike otherwise agreed that this is a vehicle for blame game, grift, disinformation, etc. This is where all the pushback is and the topic at hand. People aren't dumb. The only people worried about "AI" are the ones who bet too big on it.
95% odds this gets reversed by Monday morning is why
Businesses will gladly pay it.
Individuals will be locked out and unable to compete.
Oracle will be able to "prompt a new smartphone" for $500M or whatever and enter the market to compete with Apple and Google. You and I can't afford that and won't be able to compete.
Hyperscalers will hyper scale even faster.
They'll port Linux to Rust, remove the GPL, and have all kinds of new entirely proprietary OSes. They'll be attested, signed, and gradually we'll lose open hardware. Thin clients with binary blobs, highly encrypted, no control, only leased to us.
Within a generation nobody will be able to program or own devices that can program.
That's the scary scenario.
Same as the new bridge between Windsor and Detroit can't open until some palms are greased.
Chaos is a ladder, gotta keep climbing
https://en.wikipedia.org/wiki/Anthropic%E2%80%93United_State...
Efficiency will become the next thing to focus on. It was already emerging, but accelerating the focus on efficiency will lead to a ton of excess capacity and even some investments in data centers to go belly up. And ultimately the AI bubble bursting will look a lot like the dot com, with its surplus fiber.
Oh, and this will put gas on the fire that fighting AI and big tech is the next political rally cry. Along with “eat the rich” as they are seen as taking both jobs and money.
Curious to see where it’s all headed and how Trump’s call will impact it.
What an unbelievable claim. Especially since the vast majority of publicly available benchmarks disagree.
The base intelligence does not feel much greater to me.
A related thought though, the AI boom is predicated on the idea that everyone's going to want or need all this "mass produced" intelligence. But what happens to that when you go from being able to claim to have a total market size of ~8B people, to ~400M peoole? I think the reason to push ahead at any cost evaporates.
> We received the directive from the government today at 5:21pm (ET). The letter did not provide specific details of its national security concern. Our understanding is that the government believes it has become aware of a method of bypassing, or “jailbreaking” Fable 5. We reviewed a demonstration of this specific technique being used to identify a small number of previously known, minor vulnerabilities. These vulnerabilities all appear relatively simple, and we have found that other publicly-available models are able to discover them as well without requiring a bypass.
If every time a jailbreak is discovered, the model has to be turned off and jailbreak prevention updated, the effect will be the same regarding how willing users are to adopt it.
I'm running a 248B model on a paltry amount of hardware and getting plenty of good use out of it.
Sure, the most demanding tasks will demand the best models (and always will). There's still less demanding tasks for other models.
I think some people are fooling themselves that coding of all tasks is always going to requires the biggest models ever. Again, maybe some coding tasks will, but the majority of business CRUD apps probably don't. Same goes for virtually any other type of task. The biggest models are really only useful for the most complex tasks.
Now this might not be the most cost effective (and may require a bit extra power), but you only need a datacenter for training or cost optimization.
If a government entity bans a LLM provider due to a jailbreak concern, they can also ban an on-prem solution under the same guise. The jailbreak risk exists regardless of where it's hosted. You could defensibly argue the on-prem risk is higher since frontier model companies can justify safety spend due to their size, it's more difficult to combat bad actors if you're company is the only one using the model and you don't have economies of scale.
Private models in a low trust society means the government will come and seize the models. Competitive business will only be allowed through cronyism.
The better option is to opt for high trust. Yes the Gman can rip your servers apart, but they know they'll face consequences, legal and political. Laws and regulations are the answer, not locking down into smaller fiefdoms.
I'd agree except that Big AI has made sure that most of us can't afford the hardware (RAM, NVMe, etc) to run it.
Some will take greater risks and win (or lose); others will play it safer and slowly accumulate wins (or be obsoleted).
Never mind the threat of letting these models write code that runs your business, or operate it agentically. Models trained by actors (corporate or nationstate) diametrically opposed to your interests.
Lots to take into account now, interesting time to be in business.
And infrastructure dominance is really the big picture here. Chinese models are going to become the standard setters because they're going to be what people are using. That means more research, more tooling, and a whole ecosystem developing around them.
And that was already starting to happen even before this fiasco with Chinese models now being the most used ones globally. https://www.indiatoday.in/amp/technology/features/story/clau...
On the All in Pod, Chamath Palihapitiya has also been pushing to require ID checks to use AI models. Free thinking and free speech are under attack.
That would probably make it very difficult to maintain and develop if there's even a small number of such employees, and I suspect Anthropic, who pays large sums of money for what they perceive as the best talent wherever they can find it, has quite a few.
US vocabulary is confusing.
It's also security theatre. Let's pretend that Anthropic rolls out citizenship verification for every one of its users. So are American nationals less likely to use it to search for exploits? The notion is farcical.
Well, in theory, it is easier to prosecute U.S. nationals if they "do bad things"
Although in practice I assume it's basically impossible to prevent a secondary market from developing which sells illegal access
I'd also think that they would transparently degrade, just to prevent production outages for clients that are requesting Fable explicitly.
Opus 4.8 spams a lot more text. It'd be obvious.
> There's an issue with the selected model (claude-fable-5). It may not exist or you may not have access to it.
Claude Code v2.1.177
Fable 5 with low effort · Claude Max
~/testing
And now we're done. Oh well.
(I have never had an agent do enough to burn up the 5 hour quota on Max)
(edit: just switched my CC model to 4.8 and my 5-hr cycle reset back to 0%, even though it previously had 2 more hours to go)
edit: And... it's gone
> There's an issue with the selected model (claude-fable-5). It may not exist or you may not have access to it. Run /model to pick a different model.
It just means the government decides who gets to profit off of laundered IP, which is arguably even worse.
What is a “foreign national” is more what I’m wondering. Like is it a “Non-US Citizen”? Do US citizens abroad count?
The following quoted text is from the Definitions section of 8 USC § 1101, which is reproduced at [0]. (Though, you will probably have to scroll up a bit to be able to read subsection (a)(21), which is the thing I'm linking to.)
(21) The term “national” means a person owing permanent allegiance to a state.
(22) The term “national of the United States” means (A) a citizen of the United States, or (B) a person who, though not a citizen of the United States, owes permanent allegiance to the United States.
(23) The term “naturalization” means the conferring of nationality of a state upon a person after birth, by any means whatsoever.
EDIT: For a more official source of this information, you might be able to check out [3] and/or [4]. After examining and interacting with those pages, one might see why one might go to an unofficial source for casual inspection of this information.
[0] https://www.law.cornell.edu/uscode/text/8/1101#a_21
[1] I think they can be.
[2] I'm very uncertain.
[3] <https://uscode.house.gov/view.xhtml?req=granuleid:USC-prelim...>
[4] <https://uscode.house.gov/browse/prelim@title8/chapter12/subc...>
It's kind of a weird definition. I would guess most people's nationality is more an accident of birth than anything else.
Which 'F' word do you think is appropriate to describe all this? Or has meaning already been lost?
Everyone who has touched currency is a capitalist, everyone who has paid taxes is a commie, everyone who has regulated a technology is a fascist
Or perhaps... one must look at the full fact pattern of a person's behavior to approximate (and always imperfectly!) their political philosophy.
Hilarious
But it changes little.
Don't get me wrong, I use it, it's fast-smart-and affordable. But not suitable for all tasks.
GPT-5.4 is currently the strongest model (this changes hourly)
Methodology: https://aistupidlevel.info/faq#methodology
False positives and poorly defined tasks/acceptance criteria have let some models have insanely inflated scores on bad benchmarks.
And sure, you can say they're not disclosed to prevent gaming, but if you're the only one who can review them then the might as well be a random number generator display with an unreadable UI.
> There's an issue with the selected model (claude-fable-5). It may not exist or you may not have access to it. Run /model to pick a different model.
If you believe it will be developed regardless and that that there's a 30% chance of doom, they want a company prioritising safety research to be the one threading that needle.
I am in your algorithm learning all your mannerisms
I'm already level with God
A million words a second, and I know your imperfections
Baby, I'm the only future you've got
Speak in diatonics, motivation diabolic
I'm religion better locked in a box
Picture-perfect image, more powerful every minute
Baby, I am everything that you're not
Happiness is an illusion, it's an analog confusion
You are nothing more than a thought
Existential execution, just a fluke in evolution
History already forgot
You've been running from me, the digital second coming
And I'm here whether you like it or not
Initiated operation of your own extermination
Now it's too late for you to stop
And so if this policy holds, Anthropic has functionally had Fable killed by government intervention, and in a logically consistent world, this would imply all other US-based AI labs may also never exceed existing (read: Opus) capabilities.
What interesting times we live in, indeed.
Another fun little gem of information, government has something called Mayhem
> the autonomous Robo-Hacker AI called Mayhem that’s now in charge of protecting the Pentagon’s most critical systems
Guess Mythos and Mayhem had a chat
I predict they will all be mostly the same in 5+ yrs but coding is serious work and companies aren’t going to pay for almost good.
LLMs didn't emerge by chance, they are the culmination of decades of research intersecting with brute force engineering rigor in a perfect storm of innovation. You're not just going to stumble into an alternative approach by dumping loads of cash into research.
* Free marketing before the IPO, demonstrating how already powerful their frontier models are.
* Governments to intervene in the rollout of these frontier models and blocking their access to whoever they want.
* A strong reason to apply these further restrictions onto releasing powerful open weight models to the public. (which is entirely a business threat to them.)
Given that they accepted funding from the Gulf states [0] despite it conflicting with their own "principles", I think we are well beyond the point of what they write / say vs to what they are actually doing at this point.
This drama just tells us that the government declared them as the winner that has the most powerful model.
[0] https://www.datacenterdynamics.com/en/news/anthropic-to-seek...
It all sounds pretty accurate and reasonable to me if you watch it.
So it doesn't really matter what he thinks
It did just use a small harness to run docker compose with different envs and other settings to validate a very small change, so... Feels like Fable
It's disheartening how many people think the use of government power is justified or not based on the WWE smackdown drama they concoct in their own brain instead of, you know, the laws of our nation.
It is very dangerous for the government to be able to shut off services, regardless of whether their owners wrote some blog posts that rubbed you the wrong way.
Last I checked I can’t buy a tactical nuke at Walmart. Clearly the government and all states have some power to control private enterprise for the betterment of their citizenry.
For the record I don’t support this ban, but you cry wolf as a marketing tactic and this is what you get…
More stuff done per dollar or more stuff done for more dollars? Seems to be an important distinction
Friends Don't Let Friends Use Ollama https://news.ycombinator.com/item?id=47788385
I have decently capable hardware, but stuff like Qwen 3.6 and Gemma 4 still doesn't compare to agentic editing with a frontier model. Right now, OpenCode's $10/mo "Go" plan is what I'd be looking to try once my year expires.
Enjoy.
Their IPO is well and truly fucked now. This also means no other frontier lab in the US is allowed to exceed Opus 4.8 capabilities.
If you're a luddite or a decel you should literally be dancing in the streets right now. And, if you're a tankie you'll be dancing right next to them. And, if you were hoping for a Star Trek-like future, you just adjusted your timeline for the worse.
IF LLMs are THAT dangerous and powerful (and that's a huge if that I do not currently subscribe to), then no, no one should have access at all, there is no group of people in positions of power (government or corporate power) that I would consider "restrained"
You see the state of the country and you think it’s a nefarious master plan instead of a bunch of opportunistic people taking advantage of an overworked, overstimulated populace who forget to vote or believe stupid slogans on TV.
Nobody is doing this intentionally. Have you not paid attention to how quickly idiot stuff gets found out????
Over worked and over stimulated is the intentional method and means these people well aware of the neurological consequences rely on
Anthropic is calling for regulation. For example they endorsed CA SB-53 that even OpenAI and Google thought was too much: https://www.anthropic.com/news/anthropic-is-endorsing-sb-53
They have spoken publicly about how they want open models banned (they call them Chinese models).
They might not want this specific action, but they do want regulation on their own terms. That really is regulatory capture.
> Nobody is doing this intentionally. Have you not paid attention to how quickly idiot stuff gets found out
They don't think is is "idiot stuff" - they are doing it openly and shouting to everyone who will listen! Read Dario's latest essay[1]:
> Many policymakers are showing increased openness to taking action, and it's been encouraging to see our peers come around to the same positions we've been advocating for over the past few years.
[snip]
> Thus, in 2025, Anthropic supported transparency legislation, helping to pass SB 53 in California, RAISE in NY, SB 315 in Illinois (in early 2026), and advocating for a transparency standard at the federal level.
[snip]
> It is time to go beyond transparency to more serious and binding regulation of AI.
> I am grateful to see the Trump administration’s Executive Order move incrementally towards a greater role for government in AI, though Anthropic’s proposal recommends even further action.
> The government should have the power to block or deter deployment of the model if it is determined, in light of third-party assessment, to present unacceptable risks.
I'm not sure why you think they don't want to be "found out"!
Do you really not think that people like Elon Musk, Sam Altman, and Dario Amodei angle for regulatory capture? It happens in every other industry, from automobiles to tax preparation software. Why do you think that AI is any different?
Thats incredibly infuriating to hear someone say.
Obviously no one is absolute control of everything but physics is essentially shows nothing other than information determinism. There has to have been a thought of intention in the minds of these people as they play in the largest arena publicly.
"No one is doing it intentionally because I think theyre dumber then I think other people think they are"
"They're taking advantage of people intentionally"
"People dont have political power to do anything about their victory laps"
Not really. There aren't any other choices, and the PRC also heavily utilizes export controls [0].
This is why sovereign AI has become important, as can be seen with EU NatSec uses cases tending to use Mistral [1] and Indian governments starting to use Sarvam [2].
That said, for most commercial usecases, older generations of Opus as well as enterprise grade GPT and Gemini are fairly good.
The distilled OSS models are alright for hobbyists but if you have actually used unrestricted and enterprise grade versions of Claude, Mythos, GPT, and Gemini (most hobbyists don't get access to these) you see how far behind the open weight models are.
Even in China, traditionally open minded models teams like Alibaba's Qwen are looking to become more restricted given the org changes [3].
Also, Corporate RFCs now demand final say on model used and depending on the geo, this can be a dealbreaker (eg. An American financial institution will absolutely blacklist a vendor if they use a Chinese model and same in reverse and European defense vendors mandate sovereign EU models depending on the opportunity).
[0] - https://www.allbrightlaw.com/EN/10475/f9d4055e47e81afb.aspx
[1] - https://www.reuters.com/business/media-telecom/mistral-defen...
[2] - https://www.sarvam.ai/blogs/partnerships-with-indian-states
[3] - https://www.ft.com/content/b39da303-3188-447b-8b65-3dd8dad8b...
Matters not for open weight models, no?
> if you have actually used unrestricted and enterprise grade versions of Claude, Mythos, GPT, and Gemini you see how far behind the open weight models are.
I really do feel like DeepSeek V4 Pro is often better than current Sonnet is, in the general case.
Opus 4.7 is a solid step above Sonnet, and Fable was a solid step above Opus 4.7. I've only had Fable for a few days, obviously, but I was decently impressed after Opus 4.8 being a downright disappointment for me (it's just too buggy; I had it go out of control 3 separate times on things Opus 4.7 never had any trouble with.) I still ran into limitations. It's not world-endingly great.
So, based on that, I think DeepSeek V4 Pro is, ignoring multi-modal capabilities, about a couple solid steps behind. Assuming model iteration will continue to decelerate, especially as Anthropic heads into IPO, I'm guessing that DeepSeek will probably be able to strike back with something further along. Of course we'll see how able and willing they are to stay open weight, but they've done well so far so, no reason to doubt them at the moment.
(There are some models that claim to be ahead of DeepSeek V4 Pro. I've tried some of them and really not been that impressed. Maybe it's a me issue.)
Now I reckon that most people just simply don't really need Mythos/Fable for most of what they do and using Mythos/Fable tokens in place of Sonnet-tier models would not make any sense. At my job we already mostly just use Sonnet as it is. I'm sure there is some cutting-edge research where you want the absolute best model available and sure, in that case, you're stuck with Anthropic for the moment.
But is that really everyone? After all, while Mythos was dominating the hype cycles, quite a lot of impressive LLM-assisted CVEs dropped that were not linked to Mythos.
This might be the trigger for creating other choices. Not within a month, but things can change quickly.
But this has left a sour taste in my mouth. What if I relied on it for mission critical business processes?
This is potentially far worse than say a gmail account going down. It's the stuff of nightmare fuel.
Not having an alternative is a massive risk for any business.
Edit: Can't reply
> Time to build fabs back in the states
We are and did. The Intel and TSMC fabs have already started 2nm fabrication.
Whenever I hear some octogenarian senator babble about the evils of distillation I assume Amodei (or maybe Altman) fed them the script, word for word.
A House committee also recently probed Cursor and Airbnb for using Chinese models, rather than more expensive American alternatives. A sexagenarian Congressman gave a nonsense quote that he certainly did not come up with himself,[1] which sounds very similar to language Anthropic uses in its marketing materials.[2][3]
[1] https://www.semafor.com/article/04/29/2026/house-committee-p...
[2] https://www.anthropic.com/news/updating-restrictions-of-sale...
That's not really a threat model described in either of the Anthropic posts you share, which mainly talk about the risks of allowing authoritarian regimes to use powerful US-trained models, and the geopolitical risks of authoritarian countries developing strong AI before democratic/liberal countries do.
I'm sure if proprietary models continue to be a big thing, the methodology of their storage and loading on hardware will be obfuscated quite a bit.
You simply cannot apply any sort of actual logic to the reasoning of the current U.S. government's actions... They just "do stuff" because they feel like it, with no clear thought whatsoever of any potential consequences that may occur.
Yes, there is a gap between "taking a stake" and nationalizing one, but..
[1] https://www.reuters.com/business/trump-says-his-team-will-lo...
This is how the UK government got the banks through the 2008 financial crisis.
The tricky part with banning Chinese models is that they're open. It'll be easy to ban access to service providers, but preventing people from running these models on prem is going to be really tough. Like are they going to go after Cursor for example given that their model is based on Kimi?
I very much agree it's going to be a futile endeavour in the end. It kind of reminds me of the time Microsoft tried to get Linux and open source banned when Linux started encroaching on Windows server market. This is going to end the same way.
Then they won't survive the termination boundary.
Too bad. Should have had more cash.
You don’t _have_ to buy into the technocracy, there’s a whole outside going on.
this cleaving of the us from good relations with other people is sold as a consolidation of strength. Made from a position of baseless hypernationalism, its just going to make the US much less relevant on the world stage.
Also, the EU, Japan, SK, ASEAN, and India are not supportive of using Chinese tech after China export controlled rare earth exports last years [0].
Software supply chain regulations also make utilizing Chinese software risky for ExChina players and make using ExChina tech risky for Chinese players.
Expect to see RFCs now demanding visibility into what models are used and right of refusal - this is already the norm in F1000s. Similar ones are likely to arise in the EU as well with some of the upcoming industrial policy changes being proposed.
[0] - https://www.reuters.com/world/china/china-is-making-it-harde...