So isn’t the only logical conclusion that we have reached the max of model capabilities that the US allows to be made available to the public? Why invest in smarter models with this precedent?
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?
It seems more likely that the logical conclusion is the executive branch is mad at Anthropic, and lashing out at them with any convenient tool that they have.
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
Marketing is never ever to blame. Remember a few months ago when the U.S. government labelled them a supply chain risk? What eventually happened was that a federal judge issued a temporary injunction while calling it a "classic First Amendment retaliation." The Constitution protects such marketing; the government is not allowed to be maddened by such marketing.
Regardless of motivations for this case, I take this as a clear sign that we've reached the point where the federal government decides its within their authority to mandate this and its reasonable for them to do it.
Governments don't give up power, and once there's precedence for use of that power they'll continue to do it and begin eyeing the next power they can claim.
But it was Anthropic initiative to limit the deployment to restricted groups, it's great to see the gov following their analysis. AIs are too powerful and dangerous they should be limited to a very restrained individuals isn't?
They should have just called it Opus 5.1 and released it like normal. All this fanfare, under this corrupt regime, after they declared you a supply chain risk… Wario has horrendously bad judgment.
> AIs are too powerful and dangerous they should be limited to a very restrained individuals isn't?
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"
If its so "powerful" that it's this kind of issue, why does it even matter who "has it" or not? Like what does this mean to you? The super powerful, super intelligent AI is going to have arbitrary loyalty with one person or another?
I would be okay with that if it actually meant that. Very restrained individuals in reality would see nothing and very unrestrained governments would have access.
Are you asking me to provide evidence that in this specific instance this is an instance of weaponization of process, or are you asking me for evidence that this particular executive has lost the benefit of the doubt when it comes to weaponization claims, or are you asking for evidence that the executive is hostile to Anthropic?
Are you asking me to provide evidence that in this specific instance this is an instance of weaponization of process, or are you asking me for evidence that this particular executive has lost the benefit of the doubt when it comes to weaponization claims, or are you asking for evidence that the executive is hostile to Anthropic?
This government has proved time and time again it does not deserve the presumption of regularity and that it is more than capable of acting in arbitrary and capricious manner for petty reasons.
Anthropic was designated a "supply chain risk" despite this (a) being an absolutely absurd classification and (b) being completely at odds with the continued usage of any Anthropic products within the US government: https://thezvi.substack.com/p/anthropic-officially-arbitrari...
From that, we can very reasonably conclude that the US government has a specific vendetta against Anthropic in particular, and that this vendetta has nothing to do with the technical merits of their product.
To my knowledge, they have yet to drop that classification, despite heavy court opposition.
Additionally: technical benchmarks suggest that the most recent ChatGPT models are within maybe 10% of Fable 5's capabilities, so this being a pure "capabilities" concern seems unlikely.
Uncertainty: It's possible that we have just suddenly reached the end of public AI releases, though - if ChatGPT 5.6 also gets blocked, that would be very good evidence of a general, non-weaponized policy. Given the recent Executive Order requiring pre-release audits of frontier models, this is somewhat more likely than it was a couple weeks ago.
I still think things add up to "weaponization is the most likely theory" and that one is being disingenuous to dismiss it as a reasonable possibility. But it's certainly NOT the only reasonable possibility.
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 ban on exporting cryptography in the 90's lasted for years, and got to be a major pain in the arse for the entire web industry in its early years. The US govt can be very stubborn about this stuff when it wants to be.
Almost all of the major tech companies are either HQ'ed in the US or have a very significant US entity, and make up probably about half of the S&P500. The US's power has changed and is actively changing, but the US still holds all the cards in 2026.
So, the last time an AI-related export control was imposed - NVidia chips deemed to be "too powerful" - how do you think that will work out? If the US is holding all the cards, why is China now refusing their chips?
Don't you think these types of restrictions will weaken rather than empower the US?
I think culture moves a lot faster than you believe.
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.
> So isn’t the only logical conclusion that we have reached the max of model capabilities that the US allows to be made available to the public? Why invest in smarter models with this precedent?
95% odds this gets reversed by Monday morning is why
I realize these models are locked up pretty tight and terabytes in size, but in a future like that, I don’t see them not being leaked via an insider. The weights have to be loaded into VRAM at some point.
Every hyperscaler hosting these models outside of FEDRAMP environments has been compromised by every regional power’s intelligence services. Fable was running all over the world until today.
AWS and friends are very good at providing excellent enterprise grade security, but it’s literal child’s play for nation state threat actors to exfil these models.
TEMPEST / EMSEC alone is a wide open door for unclassified datacenters when the Mossad’s out to get you.
I'm skeptical that you're going to be able to reliably exfiltrate ~10TB of model weights using TEMPEST. Which is not to say weights are secure, just that this isn't the threat model I would be concerned about.
That would depend on what gets leaked, as I'm not so sure that the weights by themselves would be enough to replicate the architecture. I imagine some part of the secret sauce will remain in the architecture, and the tensor dimensions may not be enough to decode it.
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.
But you can see this is not true (yet); competitors/Chinese labs are less than 6 months behind: either via leaks or by just stumbling on the same improvements with time/effort.
What chinese labs are on par with GPT-5.3 and Sonnet 4.6 that I can go and use today? (granted they're 4 months ago, not 6 but nothing was released in Dec/Jan so I rounded up).
I don't think it's a good idea to give the crowds that kind of weapon. The first thing they'd do is "liberate" the model aka remove guardrails and safetly-protocols and brag on X / reddit with it and throw it into the public. That's only cool for a geek that doesn't think about the ethical impact of such a move. You'd basically become responsible for anything that is done with it, forever - have a good sleep. /s
As opposed to what, the US military, or better yet Israel (because we all know they won't be excluded) using that model to drive weaponry that kills people?
Your hypothetical implies that there is a better alternative, but when those models are "restricted", in practice that means that the only people who have access to them are precisely those who can and will use them for the worst kind of shit. So yes, releasing them to the public is a better deal, ethically speaking, at least then the playing field will be slightly more equal.
There are plenty of weapons (see custom made virus) which no state actor (or even an informal militia) would want to release, as these weapons attack everyone. But, open access to details of its construction leaves everyone vulnerable to motivations of small groups of crazy individuals.
What if I told you there are no safety guardrails. I used GLM 5.1 and had fable literally build a harness to avoid triggering guard rails.
I built skills carefully and had Fable doing vuln research and exploit repro in a few hours. I called the project manhattan. The GLM models are down for almost anything so I named it Oppenheimer. It orchestrated the fable CLI agents via tmux. This whole Fable/Mythos thing is such a fucking joke. It is all PR and theatre and they know it.
I’ve been doing pentesting with LLMs for a while and only hit a few “nope I won’t do that” and one “this conversation is flagged for being against the TOS”. No idea what the guardrails are but they are trivially abused
If the frontier models will take as much money to train as they do now, there is no way the wealthy are able to afford their training just for their own consumption. Financing of this whole thing rests on the models being available to companies and consumers who are willing to pay astronomical (compared to other software) sums for it.
I can very easily see a licensing requirement coming soon. Running a higher-grade AI will require a govt-issued license, which involves a six-month application process, explanations of why you need to run it, where it's going to be stored and who will have access to it, pretty much the same as non-USA countries deal with firearms.
Reasoning? More customers = more revenue, there's negative financial incentive behind restricting TAM. In the same way the iPhone isn't restricted to only the wealthy, that would significantly reduce total goods sold and thereby revenue. In addition that creates lower economies of scale, lower network effects etc.
Reasoning: the poster blames all evil in the world on "capitalism", "corporations", and "the rich". The aforementioned are conspiring to gatekeep us all from the obvious good of poor, communist anarchy.
Pay $1,000,000 per business function you want to build.
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.
Very few businesses can pay 1M without blinking. None of my customers. But yeah, I got the gist of it. Incumbents like moats and happily pay money to build them. Note that the pricing of Anthropic's models usually increases for new models. Chinese models cost 10 or 100 times less. Are they less capable? Maybe, but they are alternatives unless credit card companies start banning payments to them.
People have to eat food so they will keep doing business no matter what. If AI cost too much, they will do it without AI. Any resource that costs too much is replaced with cheaper alternatives. AI is no exception. At worst most of the IT business will die and we will make money doing something else.
That's the fundamental difference with open models, anybody can run and tune them any way they like. The real difference in philosophy is that Americans companies treat the model as the product, while Chinese companies see models at infrastructure you build products on top of. You amortize the cost of deploying it at scale by sharing knowledge and iterating quickly to bring the cost down.
The logical conclusion is that the US administration has decided to run the country like a robber baron and is demanding bribes from AI companies. The only question is whether EU and China can effectively attract American researchers.
For the sake of argument, assume everyone is working on good faith and at least believes and means the things they're saying.
The US government believes that Fable/Mythos is a weapon that needs to be export-controlled, and limited to only US customers. Presumably OpenAI/xAI/Google would face the same constraints, for the same reasons.
OS/foreign models are unaffected - OS because they cannot control who runs them, and foreign because they are not controlled by the US government. We could assume that China will implement the same policy controls, but they see the world differently so might not.
So US AI companies are then limited to the US market, effectively, after about six months (the lag between the current frontier models and the OS models). They have much less incentive to push the envelope to create better models, because the US govt might also ban those completely.
The investor froth around the race to AGI dies, so valuations shrink (the current IPOs may be affected), and presumably the bubble bursts. None of the AI companies can afford to continue building data centres, so that all dies immediately. US GDP drops by ~5% because of that alone.
In a year's time, the US is in a major recession because it gambled so hard on AI. Europe less so, only because it was such a distant follower in that race. China is more-or-less unaffected. The best models are now OS/foreign, and AI is moving forward more slowly, but still moving forward.
Get to Tuesday, restriction is lifted. Get to Friday, restriction is back on. Confusion reigns.
Interesting scenario though: do the other labs attempt to “dodge” the import restrictions by claiming their models are “dumber and not a threat” thereby maintaining larger market access.
If so, doesn’t this basically force a stall in US-based development? EU will keep doing its thing at its pace. Chinese models will get a boatload more popular, but will probably slow down as they can drop to whatever pace they wish.
cynical follow-up: if they’ve plateaued, is this a clever way to avoid the negative consequences of the market implosion that a substantially substandard model release would cause, thereby giving them an “out”?
A Holmes indeed... your deductive powers are piercingly perceptive! (the event chain was a joy to follow, gave me ai2027 vibes, but slowdown like)
Of course, the world is not filled with rational actors, and the probability of the current administration allowing the market to tank like that seems next to null, so Occam's razor (or whatever) would point to another TACO inevitably incoming
I'd certainly bet on your scenario if it was reasonable to assume the US and China could get over the 'race to the top or die at the bottom' dynamic
so far ai2027 seems to be playing out to an almost uncanny accuracy, realpolitik obliterates the façade yet again
If the major nations that host companies that create those OS models implement export control on top models, there won't be any new OS models with top capabilities.
Assuming anyone involved in this crap is operating in good faith is foolish at best. The only thing any of them give a shit about is accumulating money and power.
Agreed. Going by patterns in the Iran war, members of Trump's family/in-crowd will invest in AI while it suffers from this decision, and then 15 mins later Trump will reverse the decision.
The thing is, that blatant market manipulation is playing with fire here, as so much of the US economy is invested in the AI bubble.
>> Why invest in smarter models with this precedent?
The public may not see more improvements but I'm sure they government will be forcing AI companies to continue improving them for their own use. Those schools aren't going to bomb themselves.
I don't know, I've been using Mythos this week quite sceptically and I found it to be incredibly dumb. For instance gave it a dialogue between 3 people and it was constantly mixing up who said what to whom, which looked like early Gemini behaviour. But latest Opus does that too. It would also make nonsensical inference about given papers and only correct itself when pointed out what it said wrong. If that is what US government fears... maybe the fear is that someone follows the dumb things the model suggests.
it feels like it's mostly just tuned to up it's level of capability on long horizon tasks - stop context rot and keep persisting at all costs until a goal is done.
The base intelligence does not feel much greater to me.
This is a ridiculous thing to test on it. Other models are trained on that kind of thing, use those instead.
Fable was designed for _really_ hard software engineering problems. Possibly large, but especially hard. For those tasks, you feel the difference immediately.
No it wasn't, Fable is a general purpose model for use in regular chat, analysis, as well as coding.
And yes, the parent poster is accurate, Fable is just as prone to moronic mistakes as Opus was. Stop being so AI-pilled.
Codex is still a better model, and yes, for the hardest engineering problems. I use Claude for UI/GUIs and Codex for all my backend, because I have 20 years of experience of actual hard engineering, and I can see that Codex writes, cleaner code, and is far more steerable.
Bad engineers think Claude is better because it writes more lines of code and is more "proactive", but lines of code doesn't make a better system.
> Fable is a general purpose model for use in regular chat, analysis, as well as coding
This is a forum filled with experts. Putting marketing aside, in a forum like this, it is most useful to assess models according to the toughest problems in the domain they were specifically refined on. For DeepSeek, that's math. For Claude, that's programming. Gemini and ChatGPT are generalist. Yes, you can use every model for anything you like. But Fable is a bit special, it's very expensive, and very clearly designed for particular types of tasks.
> Fable is just as prone to moronic mistakes as Opus was.
"Just as" is up for debate, but yes, all models are capable of moronic mistakes. That's not helpful information though.
> Codex is still a better model
You're comparing agentic workflows, which relies on a lot more than just the underlying model. It sounds like you're using it like a precision instrument, which is great! It's very different compared to my use cases though, and the ones that Fable seems to excel at. I'm using it for scientific computing, and you really, really want it to one shot a solution. It's either the right algorithm for the task, or the wrong one. So for the hardest problems, it needs to successfully implement a solution in effectively one shot. I use Codex too, but it's often too careless for the delicate tasks. If it gets it wrong, it is really hard to steer it back. You have to start from scratch.
> Bad engineers think Claude is better because it writes more lines of code and is more "proactive".
Think you missed the mark on this one. Not really an engineer, have as much experience as you do in my job. A solution to my problems comprises few lines of code. Fable actually gets it right, first time, every time (so far), but this is with a very long prompt and a bunch of attachments. No other model has done this for me. Not shilling for Anthropic, just impressed. This isn't particularly subjective for me; it is quantitatively measurable.
Don't assume everyone using AI is going to have the same experience you have, or the same types of use cases. And please don't assume that because others have different experiences that it makes them "bad".
Also, Claude has always been mediocre at creative tasks. For your line of work, I would have already recommended Codex hands down.
I tested it on that too. A problem I usually give a model to test is to optimise already well optimised function that performs certain calculations. I give it reference to CPU instruction set, how instructions can be paired to take advantage of superscalar execution pipeline etc. In that test also it fell on its face by producing code that was demonstrably slower and with extra bug.
There may be a temporary plateau. And it could have fascinating macroeconomic impacts.
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.
You touched a great point, I wonder how the markets would react to this. No wonder the order was released just after the end of business on a Friday. Let me guess, something will happen right before market opening on Monday and some bets will pay great dividends.
If there’s one thing that’s certain it’s that Trump will do something just after markets close on Friday.
But I hadn’t considered this fell into that category. Except maybe as a direction from Iran. You make a good point, it may trigger immediate reactions in the market. Not just 3-6 month ones.
I wonder what the counterbalance will be by Monday morning.
This is utterly daft to say if you actually used the thing for hard problems, something that benchmarks have been known to be unable to capture. It is night and day compared to Opus and every other model out there. It was nice while it lasted.
It's strange how uninformed people are when they are so willing to to make assertions. I used it too and it really felt like a generational shift and not an incremental one.
These threads about Anthropic always seem so astroturfed with some of the loudest and most uninformed people around.
It's really annoying when people seem to assume that their experiences and use-cases are universal, and that other experiences are "psychosis". Fable seems to excel at tasks that are research-level hard. The vast majority of people do not have these problems. FAANG engineers probably don't either. Codex is great, I use it too. It has a major advantage in terms of token limits so you can often brute-force things through. If it is a task that Codex can do with enough tokens, it isn't worth using Fable for it. I used Fable on the tasks that Codex and Opus consistently failed at. Delicate computational geometry tasks, mostly, really darn hard ones. It worked first try, every time, every single one. That's terrifying to me. I don't know what to tell you, I don't care about hyping Anthropic.
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?