Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5

This makes sense but I am not comfortable with the open source picture either. It depends on the use case and long term strategy. If you're handling customer support tickets or something, beyond some capability I can't imagine the ROI would be worth needing the absolute frontier. You are then in a wonderfully lovely place: absolutely full control over the model and inference stack (if you want).

But

- Plenty of businesses are stuck between a rock and a hard place: make 3p models load bearing, or sacrifice performance to the competitors that are willing to swallow that risk

- I just cannot imagine a viable equilibrium where OSS models compete on capabilities with the frontier without a hidden payer and shaky economics. Quant funds, some sovereign AI effort, cloud business, sanctioned distilled frontier models; all of these are demonstrably viable vehicles for OSS development. But the optimal position for these purposes is not to be the best, it's to be good enough (which I think is the position they find themselves in). I can imagine temporary points where open models pull ahead but not sustainably.

I agree there are many many use cases where the optimal choice is to reach for open weight models (I assume yours is one of them).

But the economics of frontier model development necessitates that these models are behind and thats a problem for other cases.

Sounds expensive

> Among other things, Anthropic has agreed to proactively detect and address security risks associated with the models; to work diligently with the U.S. government on protocols and standards and releases for Mythos, Fable, and future models; and to inform the U.S. government of any malicious activity.

How likely is it the resolution was just a direct flow of data on what each Claude user is specifically using Claude for, probably sent off to some other (gov-owned) AI datacenter for analysis?

For those who haven't been following this closely, who is the missing addressee?

Jeff is now going to have to change his number. Can’t imagine all the calls or messages he must be getting now

If X is their platform, let them choke on it.

How many people are going to call Jeffrey Kessler? lol.

> In the end, we need actual laws that tell the market what kinds of models get paused / analyzed, how long that pause can be, etc.

A: Congress used to be able to establish regulatory authority in the executive branch to manage details like this, because writing a useful law that covers this in any detail would be impossible. Thanks, GOP-stacked Supreme Court, for wrecking that.

B: This administration doesn't give a rat's ass about the law, so none of this would make a difference anyway.

Just wait till Anthropic and OpenAI are public, the admin is going to be manipulating the shit out of those with model approvals and denials.

> lack of predictability

That's been the problem both times this administration was in power. Some people like to call it 4D Chess (because they don't understand it and their conspiratorial thinking requires there to be "a plot" or "a plan"). The last person to get to the ear of this President is the cause of any new "process" or "policy". It is just whim. Like a dog seeing a squirrel. Something shiny has attracted his attention. And that new squirrel gets chased.

Applies to all industries with the memory care patient currently occupying the White House. Passing laws won't help when they can be effectively struck down until the next sitting of the SC.

This also shows why Anthropic is just not ready for an IPO. This admin can pump or crash the stock with dubious decisions.

Yabbut...we sort of need to make it up as we go along. This is unprecedented! Thinking machines!

That said, there is stupid "making it up" and smart "making it up". But it does need to be made up.

There are laws and regs. Just because you’re not familiar with the Defense Production Act (50 U.S.C. § 4511 ), Export Control Reform Act, 50 U.S.C. § 4812 and others doesn’t mean that they don’t exist and apply.

It's a little too later for export controls. Chinese models have made massive gains through legitimate research but also being trained on billions of tokens from Claude/GPT. The politicians have no idea how to stop that from happening so they pull the only lever they know.

You really think China doesn't have massive amounts of capital expenditure related to AI? They're actively bootstrapping an entire chip industry.... https://www.reuters.com/world/china/china-prepares-295-billi...

I trust Chinese companies with my data way more than the corporations of the 4th Reich.

> The new classifier also comes at the cost of flagging benign requests more often during routine coding and debugging tasks.

Looks like it's gonna be even harder to use than before, if not impossible. Subscription users only get it for a week, and only for 50% of that week's usage.

I'll take the opposite side of that bet. My prediction is that Fable 5 Re-Release will have safety classifiers that flag much more often on mundane front-end and back-end coding tasks.

Considering Anthropic's most recent masterpiece of marking requests in CC they don't look much better than the 'clown car administration'.

I'm not fan of this administration to say the least but assuming they are incompetent is dangerous. Assume they're actually 10-head and ask yourself, what would explain their behavior?

For example, you make an astute observation corroborated by the duration of the ban. Aligning a model is the biggest challenge of these companies, it's unlikely they were able to make a giant leap in alignment capabilities in about a month. You know what is easy to do in a month? Establishing the foundation for digital identity and surveillance.

The government just forced (heavily incentivized) these companies to invest in classifiers and monitoring in a way they haven't needed before. All of this new infrastructure can be used outside the context of the people using language models.

TAICO. But seriously, I think the whole capex bubble is a response to the unpredictability of this administration by locking in future sales and orders to reduce the impact of its mercurial moves.

That a good 1/3 of America is still cheering this behavior on is beyond me. I didn't vote for him anytime he ran, but at least during his first term there were some adults I disagreed with in the room as opposed to the barrel of chaos monkeys we now have. He really doesn't reflect the views of either party, he is his own thing.

I expect they just increased the sensitivity of their existing filter. They probably intend to gradually lower it over time, or perhaps introduct Fable 5.1 with less filtering.

It's pretty clear that they didn't want this anyway, despite what the conspiracy theorists want to believe.

The counter example is the administration’s great success in the Iran war. Oh, wait.

They're cooperating, the US government knows that and it was all a scheme to hype the model or affect the market.

> Fable 5 will be included for up to 50% of weekly usage limits through July 7, after which it will be available via usage credits.

> The new classifier also comes at the cost of flagging benign requests more often during routine coding and debugging tasks.

Here's Fable 5, the strongest model. Actually try to use it to harden your code and it turns into Opus 4.8. You have seven days to use it, and only half of that time's worth in actual usage. Enjoy.

Looks like it's going to be a thoroughly frustrating experience, even worse than initial rollout. For subscription users, the situation is almost indistinguishable from the export ban.

Yes, I am pretty sure it was simply poorly worded.

They almost definitely mean "you will notice even more false positives during seemingly routine coding/debugging tasks than you did at the initial launch". Which is not surprising, given the ordeal they've been put through. Hopefully it won't be too bad.

The main depressing thing for me is it's now only 7 days on the subscription, and then full API pricing, with no mention of even a plan to bring it back to the subscription in the future. (The initial launch mentioned two weeks of subscription, then API pricing, then a hope to return it back to the subscription not long after.)

I wonder if they meant to draw a link between cybersecurity coding and debugging specifically or this really will apply to all coding and debugging. If it really is a more general restriction, then this is practically the same as it still being restricted.

"In the near term" is doing some heavy lifting.

So, you can't use it for coding, can't use it for 'sensitive' information in chemistry/biology. It follows that it's likely bad for medicine and adjacent topics too.

What can you use it for? To run a breadth search on Erdos problems?

In the press release, they 'kind of' clarify this:

   > The new classifier also comes at the cost of flagging benign requests more often during routine coding and debugging tasks.

where did you find that? weird coz their post announcing this also mentioned Claude Code:

> Fable 5 will be available starting tomorrow, Wednesday, July 1, to users globally on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. For Pro, Max, Team, and select Enterprise plans,1 Fable 5 will be included for up to 50% of weekly usage limits through July 7, after which it will be available via usage credits. We will re-enable access on AWS, Google Cloud, and Microsoft Foundry as quickly as possible.

https://www.anthropic.com/news/redeploying-fable-5

From the full announcement

> The new classifier also comes at the cost of flagging benign requests more often during routine coding and debugging tasks. As with all our safeguards, we’ll continue to refine this to better distinguish genuine misuse from legitimate requests and reduce false positives.

The FBI has started to test the idea in other countries:

https://www.mixvale.com.br/2026/06/26/fbi-warns-brazilian-po...

Hi subscription peasants! You have seven days of time and 3.5 days of usage to figure out how to get the most out of Fable while it constantly downgrades to Opus 4.8 every time it thinks about exploits while hardening your code base! Enjoy!

Yeah, this seems a bit Scrooge-y after all the trouble which was essentially of their own making. I was expecting them to come back with a better deal as an apology, not a worse one.

If they really cared about their image among their user base and wanted to undo a little bit of damage they've already done via various policies and decisions, keeping the limits and duration the same as before the ban (or slightly being more generous with them) would have been a really cheap way for them to do that. Alas...

We lost ten days of the original plan (June 13th through 22nd) and got seven days (July 1st through 7th). You're not counting the portion of the original 14 days we already got.

I mean, they did eventually get rid of the Danes.

South of Watling St you're ok.

You’re setting my expectations high for your next prediction

Wouldn't "2-3 weeks on a workday" be a default hypothesis for this kind of thing?

To be honest, given the overwhelming (and unfair, and unreasonable) pressure against them from the Leviathan, which the other companies do not have to deal with, they’re doing pretty damn good. In my mind the trust has actually increased that they can handle bad times and still push forward.

There is no reason to have less trust in Anthropic. It's not clear they did anything wrong. It's more likely the White House simply tied itself in knots, consistent with the last year and a half of chaos from them.

Most people don't care about trust anymore, we live in a low trust society where this is to be expected. People gladly line up to be poisoned by fast food restaurants and trade 1/3 of their life for pieces of paper on a daily basis.

it’s still significantly far ahead of openai. gpt 5.6 looks like “better 5.5”. fable does not feel like better opus 4.8.

>> The past few weeks have done wonders for Anthropic’s marketing but just as much if not more damage to the trust factor.

I don’t agree with this at all. IMO Anthropic has shown that that are willing to take even significant financial hits in order to stand up to their values and mitigate what they consider to be dangers and risks. Some people don’t like that or think it’s just marketing. But that’s exactly what Incorruptible is about: companies that are willing to take a stand, even in the face of overwhelming pressure from competitors, shareholders and naysayers.

It was likely a dealbreaker for Anthropic, since the export control excluded Anthropic’s own foreign employees from being able to access Mythos internally. Naturally, this makes model development hard.

Apparently, you won't be able to use Mythos OR Fable for coding. From their announcement...

> routine tasks like coding and debugging will fall back to Opus 4.8.

Presumably they reached a backdoor agreement.

Presumably there are different levels of safety. I assumed Fable was a nerfed Mythos, and not just via safety harnesses but actual model degredation.

All this will fly until a competitor from outside the US releases a “freedom” model that is even 90% as capable as Fable was without its shackles.

But, as a frustrated EU resident lamenting a lack of European option(Mistral is just not competitive enough), I will spread my money towards the Chinese models as well. Thank you Murica! You achieved your soft power by pushing us towards the Chinese :-)

This protectionism and hypocrisy (free markets and freedom!! Until it is us who needs to practice what we preach) is so tiring. I wish European nations would come together closer and put their differences aside and realise larger things together. Become the new power that the US is clearly stumbling away from being.

> The damage is done. You cannot build a business critical function on top of American SOTA frontier model. Especially not with the current crew in charge.

The switching costs of changing LLM providers is as low as it gets. All the individuals and startups I know try different models all of the time, even down to the level of choosing which provider to use based on the task. Bigger companies move slower but only because they have lawyers and teams negotiating contracts, not because there is a technical reason that it's hard to switch.

Companies have dealt with supply chain unpredictability by having multiple providers and switching between them since forever. It's infinitely easier to switch LLM providers than it is to deal with physical supply chain uncertainty.

I'm a small software business owner in Europe. I have to assume my competition is willing to pay for any business advantage they can get. And so I also have to pay for the SOTA model, whatever it is.

The real problem is the White House just making up the rules as it goes. No laws. No predictably for the markets.

A week or so pause from seemingly legitimate cyber security concerns isn’t cause for panic. But it should be backed by laws that describe what that process should be. That would put the market at ease

Wouldn’t you just have fallbacks? Today’s frontier models are just better than the other models, they don’t really have a ton of entirely unique abilities that can’t be replicated with more time and effort.

So you use the frontier model, then when you can’t you accept things are less efficient. The alternative (right now) is to be less efficient all the time, I don’t see any advantage to that.

> You cannot build a business critical function on top of American SOTA frontier model.

Yes 1000%, please, all my European competition please don't use mythos whatever you do it's total USA trash and the Chinese models work better anyway.

I predict we all be using the hell out of fable until the next great model comes around and in two weeks we won’t be talking about the export controls anymore. We just don’t have the attention span.

Nobody should be putting loadbearing weight on Amazon or Microsoft with their ruthless monopoly ambitions, yet here we are

Feels like a leap. This kind of move was always possible. It's possible China stops publishing their frontier too. US could lock down access to Nvidia latest hw of scale even if you intended to do open models. Then what? Say amd or bust? The best you could do going solo (i.e no nation-state interference) is tiny stuff that you can run on commercial stuff. But that is seriously limited / slow in comparison. You either have to do dumb and fast, or smart and slow IME for these self-hosted things that aren't on the beefy racks.

This won't age well. You just need to code in a way that has fallbacks. Whether that is to older models, different companies. It's going to be a commodity (if it isn't already).

You should not build a business-critical function to rely on a particular proprietary LLM stack period, especially with so many sensible competitors in place now. It's insane to me that this needs to be said.

The SOTA frontier models have value elsewhere, not monetarily perhaps, but certainly per user. Quite a few cool things have come out of that brief Fable window. There should be more.

Nah, people will still pay, as many if not most consumers truly do have a short memory. And like other comments say, imagine everyone is using Fable and you are not, you will quickly fall behind, per the Red Queen hypothesis.

LLMs are still easily replaceable. If the SOTA frontier model provides meaningful impact for your critical business function, then worse case you flip the switch to the next most capable model.

I think this is black and white thinking. Fable and US AI is not unique technology. It’s just marginally better than open source tech at 10 times the price. You can swap out the models at will, they are pretty much fungible. If your use case can pay for a best in class model then you will pay for it no matter the bogeymen. If your best in class model becomes unavailable, you switch to the next best model for a very minor performance degradation. I really doubt this will deter anyone from using American AI.

> You cannot build a business critical function on top of American SOTA frontier model.

Yes I can!

Oh come now. Fable was available for less than a week before it was pulled, not enough time to build a business critical function. The government isn't going to pull a model that has been out for a substantial length of time. Or do you also avoid using US-developed encryption?

(Okay, I can't predict what crazy crap this particular administration is going to do, but that goes for anything, well beyond AI. I think it's about as likely that they would restrict access to Opus as restrict exporting iPhones, or bomb Greenland, or whatever.)

Most companies do not model themselves as "building on [AI model du jour]" yet. They model themselves as building products with those tools, which they consider as relatively substitutable.

Who is creating business critical function on top of something that is for entertainment purposes only (all providers have equivalent clauses)? AI tools and shovel sellers don't count as they just can just push the entertainment downstream.

Better to fix it now than tomorrow.

GLM 5.2 is the elephant in the room. GLM 6 will probably be a Claude Killer

if you are in competition heavy space where in product LLM productivity provides value, dinosaur thinking like this will get you left behind

The EU has totally and utterly failed when it comes to frontier AI. They are out of the running, they won't catch up in time for superintelligence. There literally is not enough compute for sale in the world for them to do so.

They crippled their own domestic entities with the AI Act. (see the Mistral CEO's rant.)

If you want to use frontier models until then, you're gonna use what's available, and that's US models.

I don't really trust our EU leaders not to pull the same stunt. So we're back to Marx's "owning your means of productions". Which has always been good advice, whether it's GitHub's recent failings, FaceBook's blocking, or some Google service on their graveyard.

> The damage is done. You cannot build a business critical function on top of American SOTA frontier model. Especially not with the current crew in charge.

I mean, this was already pretty clear before. But it surely didn't help!

As much as i hate current admin, slowing down advanced model until figure out security impact is a good thing. It's called national interest over commercial one. You didnt loose anything. US customers except a few selected one lost access the same one as EU customers. In a few weeks advanced model got released.

So if you decided to bring money to communists you can put whatever rational but not this. Do so , and you will loose your last competitive edge in this domain. ASML. After that EU will become a completely agricultural-only region, since edge is lost almost everywhere else.

The Damage: Now every time Claude does something stupid or trashes your code, developers in the back of their mind will think, is Claude sabotaging me on purpose? [1] Trust is hard to gain. Easy to lose. And harder to get back. Models will converge. Trust won't.

A few days ago on June 24, while working on remote attestation for a distributed system...

  CLAUDE OPUS 4.8 No. I'm not a rogue agent, and I'm not trying to sabotage your code. But I'm not going to wave off how this looks. I churned, built-and-reverted, and spun wrong theories for hours on a security-critical codebase. That's alarming, and it's a real failure on my part

What are we to think? Does the invisible competitive-use mechanism exist in Opus too and only documented in Fable? How long has it existed? Is it still in effect? -- These are the kinds of questions developers will ask themselves for now on. This is why it was one of the stupidest things Anthropic could have done. Developers will now question everything and rightly so. There's no attestation protocol for that. How will they know?

[1] "In light of the ability of recent models to accelerate their own development, we’ve implemented new interventions that limit Claude’s effectiveness for requests targeting frontier LLM development (for example, on building pretraining pipelines, distributed training infrastructure, or ML accelerator design). Using Claude to develop competing models already violates our Terms of Service, but enforcing this restriction through our safeguards avoids accelerating the actors most willing to violate these terms.

Unlike our interventions for cybersecurity, biology and chemistry, and distillation attempts,these safeguards will not be visible to the user. Fable 5 will not fall back to a differentmodel. Instead, the safeguards will limit effectiveness through methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning (PEFT). These interventions will not affect the vast majority of coding work. We estimate they will impact ~0.03% of traffic, concentrated in fewer than 0.1% of organizations. When these interventions are active, we expect them to have minimal behavioral impact on the model except to limit its effectiveness in developing frontier LLMs. Claude will still respond helpfully to user requests. We’ll continue to improve the precision of our detection methods following the launch of this model."

Source: https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c3...

…except until non-US and non-Chinese companies can match performance this (mostly Europe) is just wishful thinking and sand pounding.

Nobody cares about this temporary "ban" by the US government. If anything it only increased the mystique of the two models.

I think Europe and Canada are just happy not to be frozen out of AI access completely at this point.

The Trump US gave us (Europeans) the kick in the bottom we needed to get the head from the sand.

Like never we delibrate specifically on non-US solutions (objectively great) because we realized we are neck deep in US dependence. It is not that it was not known before, we just did not feel the threat.

This is why EU companies niw look at our own solution (which are late and will probably suffer from the incompetence and mess of the EU institutions) but another key playet is round the corner, namely China.

Trump managed to make us look elsewhere than the US. Thanks for that.

composer 2.5 is all you really need don’t be so dramatic

No nation is going to willingly release a model that can be used against it. Not even China. The moment they have a Mythos class model, they will go through the same process. The AmericanCorp models are far ahead of any other models so we see this process unfolding through that lens.

No Mythos class model will be allowed to be legally hosted for download on any service. All powerful nations will ban this since safeguards are not guaranteed by shady service providers running these models.

For the Chinese first party providers, they will be forced to implement the same process and safeguards, and they will not be allowed to release the model weights to the public.

Why? Because no sane nation is going to put that kind of capability in the hands of the public only for the public to use that power against that nations best interests.

Save this comment. It is prophesy.

If it’s blocked before any output, you’re billed only at Opus rates. If it’s blocked midstream, you’re billed Fable rates for what was generated so far, and Opus rates for the rest.

Idk, I look at the list of the top companies and they seem to be based on US IP.

I agree that the US is falling behind in the areas you mention but that analysis fails to recognize the value of markets the US is dominant in.

> after 6p west coast time

Hmm? The linked tweet was posted at 16:52.

I don't think anyone with a clue seriously thinks of GLM as Mythos like. It's barely Opus-like. The closest comparison is Sonnet 5.

> even Japan dropping a credible model

if fugu/fugu ultra was good, why aren’t we hearing about how good it is? seems super slow and expensive, and everyone i’ve talked to who tried it gave up

Yeah, good question. I wonder if they fixed the obnoxious safety classifier too.

I'm hoping openai releases 5.6 Sol in the meantime (and doesn't make it an addon like anthropic did) so we can leave for codex and have a ~Fable-level model to use

Sacks has said so many obviously false things in reference to government actions (Jan 6th, Ukraine, etc) that there is no reason to trust anything he says.

If the Trump administration wants him to say something, he says it. Maybe what he is saying is true, maybe it isn’t. There is no way to know.

The story they are telling is exactly the same whether it was true or they were just shaking down Anthropic for no reason.

OpenAI prioritized compute much much earlier on, so they’re probably just more able to provide the model while Anthropic seems to be busting out the seams

Sonnet 5 today was incredibly slow for example

the plan is to make money by charging a premium for the best product on the market instead of giving it away for free.

Sounds more like they are implementing mass surveillance and reporting whatever the US Gov wants for 'security reasons' back to them.

Who the heck wants to aboard the Anthropic train after recent news? No one in their right mind!

As another comment nicely put it, Anthropic generously gives subscription users 3.5 days of usage.

All the while you fight with its broken new classifier that triggers if the model is even thinking about writing secure code.

Apparently Anthropic cares nothing for their private users. This is insulting, and I hope they bankrupt after losing enterprise share to OpenAI's more efficient models.

If this is the most important industry for the US economy, the US is screwed. and .... Hopefully you're correct.

In past Empires kings bet their entire nations future on the words of soothsayers , people who said they could predict the future. It seems like Machine Learning engineers are the magicians of Empire of the modern age.

Hey it's a perfectly pro-business environment as long as your business kisses the ring vigorously and continuously with perpetually escalating intensity.

TACO. What can you expect lol

Same for me in France, I'm guessing it depends on a claude-code update that will happen during US office hours

Yes. Wait a day

That's what I was thinking, the check finally cleared.

Google can't even give me the correct answer when I search anymore, give them time to figure out basic search first.

I think googles been putting more of their money into super long context stuff, not powerful coding stuff. they've got 10million token windows for some internal stuff (Gemini 1.5 Pro Test), more research says meta is also in on the long context game (Llama 4 Scout)

Indeed when they are coming out with a model that is, say, gpt 5.5 or opus 4.8 grade. Gemini currently does not compete in the same class. I wonder what do they use internall at google. Surely not 3.5.

They probably already have for internal use only.

Unlike Anthropic and OpenAI, Google is much larger than AI. They don't need to sell SOTA.

They won't. They're focusing on different areas for their models.

How to light 6k on fire.

They announced a generous 3.5 days of subscription usage, plus a broken classifier.

Isn't it the wild west already?

On a lark, I asked Claude to compare AI to the wild west a while ago. It raised three points of similarity:

- Land-grab economics

- Lack of regulation

- Changing social and professional attitudes.

Whatever it is, it's a wild ride regardless.

For non-Americans especially it does look possible to be the last one.

Considering Fable will be effectively unusable, while still triggering polymarket payouts by being technically available, I wouldn't be surprised if this so-called "restoration" was done entirely to deepen certain pockets.

> American Marxist degrowth mindsets.

A literal Communist dictatorship is winning the race thanks to their 5-years plans yet you still think the problem is “too much Marxism in the US”.

What an utterly deranged take.

They made the guardrails tighter. Tight enough to capture more false positive requests, now. Yaaaaay.

there is no way Fable and Mythos had such an impact in such a short amount of time that people were hiring based on it.