Forgot your password?
typodupeerror
United States AI Government Republicans

Anthropic 'Suspends' All Mythos and Fable Access After US Order Limiting Foreign Access (reuters.com) 51

UPDATE: Amazon CEO's Talks With U.S. Officials Triggered Crackdown on Anthropic Models.

"Anthropic said on Friday it will 'abruptly disable' its most advanced AI models for all users," reports Reuters, "after the U.S. government ordered it to suspend access to the models for foreign nationals, citing national security concerns. The company received the export control directive to suspend access to Fable 5 and Mythos 5 for all foreign nationals, without being given specific details of its national security concern, Anthropic said in a statement."

Anthropic's blog post writes that the directive applies to foreign nationals "whether inside or outside the United States, including foreign national Anthropic employees. The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance."

"Access to all other Anthropic models will not be affected." We received the directive from the government today at 5:21pm (ET)... Our understanding is that the government believes it has become aware of a method of bypassing, or "jailbreaking" Fable 5... We have not even received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result. The potential jailbreaks that have been disclosed to us are either entirely benign responses or are minor findings that provide no Mythos-specific uplift.

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 a report that we believe is the basis of the government's directive 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... We are complying with the government's legal directive and are removing access to Fable 5 and Mythos 5 for all users. However, we disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.

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. We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.

Reuters notes that Amazon's cloud unit AWS "said late on Friday that Anthropic has asked it to revoke access to the models for 'all users in all regions.'" Dean Ball, a former White House official who contributed to the AI Action Plan the administration issued in the summer of 2025, said in a post on X that the order suggests all "non-Americans" would be restricted from using Anthropic's latest models, including those based in the U.S. "This means you should expect to have to prove your citizenship to use Anthropic models," Ball said. Several key Anthropic personnel, including co-founder Chris Olah, AI researcher Andrej Karpathy and philosopher Amanda Askell, were born outside the United States.

Anthropic 'Suspends' All Mythos and Fable Access After US Order Limiting Foreign Access

Comments Filter:
  • by ihadafivedigituid ( 8391795 ) on Saturday June 13, 2026 @03:52AM (#66189748)
    Anthropic made Skynet? I think not.

    As it happens, the Chinese are capable of making their own near-frontier models, many of which they release publicly as open weights.
    • by thesandbender ( 911391 ) on Saturday June 13, 2026 @07:02AM (#66189838)
      So far most of the popular Chinese models have relied heavily on training with US Frontier models. Even as recently as Qwen 3.5 it would sometimes respond that it was ChatGPT or Gemini. Anthropic and ChatGPT even play whack a mole with accounts that are doing this and a large chunk of them have been tied to the PRC. It's a perfectly valid approach (technically) but it's against the ToS for the providers.
    • Anthropic made Skynet? I think not.

      As it happens, the Chinese are capable of making their own near-frontier models, many of which they release publicly as open weights.

      It's not hype, Anthropic is literally saying there's no actual risk they're aware of.

      What actually happened is they refused to let US Intelligence agencies use their models to perform illegal surveillance. So now the US government is illegally punishing them.

      • What actually happened is they refused to let US Intelligence agencies use their models to perform illegal surveillance. So now the US government is illegally punishing them.

        It's fine. They can turn to the courts for redress. Of course, it will be moot by the time the courts get around to it..

        Working as intended.

  • I assume once they go public, the hype will kind of die down [ig.com].

    Either that or it will become more shrill, as hot air is passed through a narrower and narrower factual space.
    • by gweihir ( 88907 )

      Either that or it will become more shrill, as hot air is passed through a narrower and narrower factual space.

      Which raises the question of whether there is an upper pressure limit of the hot air. My CS students are all sick and tired of this thing that the only type of LLM thesis I can get anybody interested in is of the "LLM failure" kind. These still go. Oh, and they do not have problems to find jobs. They all can afford to be selective and still will have something months before graduation. Of course, this is Europe and I teach all things related to IT security, but still.

  • Par for the course (Score:3, Insightful)

    by spaceman375 ( 780812 ) on Saturday June 13, 2026 @04:04AM (#66189762)

    Given that Trump fires experts and hires cronies with no qualifications for the job at hand, of course the governmental actions are heavy-handed ideological decrees with no basis in fact. Embarrassing them with the truth is a tricky response, but I applaud Anthropic's effort.

  • Whatever happened to "We have to be the first because else China will beat us!" "We can't hamper corporations with rules that will stiffen innovation!" Oh, the Ai questioned the sanity of having Trump as president... Yup, way too dangerous. Stupid left fake news biased AI! Get it out of here! ...

    whatever ...
  • by gtall ( 79522 ) on Saturday June 13, 2026 @06:14AM (#66189806)

    While based in San Francisco, they have offices all over the world. If I were them, I'd be threatening to leave the U.S. and not make an idle threat.

    • Considering that the orange shitgibbon is not above taking hostages - he did that during his first term already - it might be dangerous.

    • And if they did that, they would be arrested on the spot, their US operations seized and their models dies overnight to bans on receiving any more GPU's or contracts from any US based entity.
      • by PPH ( 736903 )

        This is why you do all of your R&D overseas. And then license it and import it as needed for US operations.

  • So release just enough of a hot model but turn it off and not actually deliver the computer intensive and costly product to your customers while sitting on subscription dollar. Sales still pumps out we have the most intense product, the 50 cal rifle of AI. The customers line up to hand over dollars for the less compute intensive model. Might be a signal the grif is up, or AI found a really concerning bug that is still unaddressed.
    • All the so called LLM's likely use the same backend technology with the same architectures driving their processing. Their differences may be the weight of the database'ed responses returned, The "morals" of the company. The decision making will never be any more on the LLM frontend than it is right now. They sold their morals to the world in a completely immoral space along with the rights to develop their backend in that direction, which seems to be better than the rest. Trump did years of marketing for
      • Re: (Score:2, Insightful)

        by Rei ( 128717 )

        so called LLM's

        LOL

        likely use the same backend technology

        LOL

        differences may be the weight

        LOL

        of the database'ed responses

        LOL

        • I guess i should say in other words ; the backend is achieving some type of reasoning, logic but the part that gives you directions to the nearest chiplote is not that part.....
  • by teg ( 97890 ) on Saturday June 13, 2026 @07:52AM (#66189868)

    Under this administration, you know the underlying cause is they didn't pay off Trump, his family, and his cronies.

    It's sickening how corrupt the US has become. Sure, one can argue it's been bad - but Trump has turned it up to 20, on a scale from 1-10 - just bulldozing through any limits one thought existed. "Want a pardon? Sure, buy some millions in Trump coin and it's yours." Or just giving himself a giant slush fund via a settlement with his underlings [youtu.be].

  • by ffkom ( 3519199 ) on Saturday June 13, 2026 @08:24AM (#66189888)
    This should teach even the most gullible of politicians around the world that single-sourcing your brain substitute from a US service is really really stupid. People will need to host their own LLMs... or at least permanently load-balance between using LLMs from N different countries. And yes, I am assuming here that human brain atrophy due to habitual LLM use will make people entirely dependent on such technology.
  • Can we be honest? (Score:5, Interesting)

    by DarthBobo ( 152187 ) on Saturday June 13, 2026 @09:12AM (#66189950)

    Can we be honest for just one minute and admit that this in unlikely to be about national security, and more likely the current administration intervening in private enterprise? They have been very clear that they want to bring Anthropic to heal by threatening their business (eg, Dept of Defense declaring them a supply chain risk). Anthropic's competitors have the direct ear of the administration and have shown themselves to be willing to act unethically to get what they want (Altman, Musk). Until we can recognize this behavior, point at it and discuss it out loud it won't end and we will continue a slide towards oligarchy.

    I used Fable for about 24 hours - it's impressive. We used it to review a system for potential security hardening and it did a really good job, but not much better than 4.8 or gpt-5.5 running for a longer time. And it wouldn't identify potential attack vectors. I can't say how useful Mythos would be in the hands of an adversary, but I can say that Fable wouldn't be a step change for anyone that can afford to burn tokens in agentic fan out pattern.

    • It's nice to allow anyone to use something so powerful or capable. But then you must remember when people think they are anonymous with no repercussions they really are not nice to each other and revert back to their underlying animal behavior.

      A small town can have nice things - a global town can not.
  • Be careful what you wish for. How could you not see this coming. Oh the good side, they probably care nothing about the AI 'escaping' - which more and more I'm thinking is actually the ideal outcome.
    • Thank you, this is exactly what I came here for:

      If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.

      It would however represent exactly the type of "please regulate the industry; we shouldn't possibly be trusted" response Anthropic advocates. Just a typically Trumpian-style instance of it.

      To wit:

      the order suggests all "non-Americans" would be restricted from using Anthropic's latest models, including those based in the U.S. "This means you should expect to have to prove your citizenship to use Anthropic models,"

      • And as pointed out in a Medium article (Paywalled, sorry), the whole Fable/Mythos release restriction is also the beginning of a class-tiered "You aren't worthy of this" regime.

        https://medium.com/@ignacio.de.gregorio.noblejas/todays-the-start-of-a-new-ai-era-you-aren-t-invited-to-e0172b6ac3b8

        Welcome to the Brave New World

  • Move the whole shebang to Israel. There's plenty of solar energy to be found there.

    Of course, they'll need a bombproof infrastructure...

  • I like the sound of that. Imagine if Anthropic moved their HQ to the EU where the bureaucracy is so complicated even if the president wanted to do something there would be 17 objections and 23 years of legal battles ahead.

    They'd be arguing if Fable 8.3 is allowed to teleport people by then.

    Jokes aside, instead of bending the knee to Orange Jesus Anthropic should seek to migrate away from the US. Just assume now that he will rig the elections somehow to win a third term...because if you're counting on h
    • Sure, it would run on those wonderful EU-made graphic cards...

      • It would run on Chinese-made graphics cards, just like it does in the US.

      • You mean the cards that only exist because of the chips in them, that are made in Taiwan with the EUV lithography equipment that for which the EU is the world's sole supplier? THOSE graphics cards? THAT is the trade war you want to start? Brilliant idea, CHUD.

  • Instead of building out data centers we could be giving everyone a computer capable of local inference. Obviously you need to show proof of citizenship to buy the computer - what could possibly go wrong!
    • by PPH ( 736903 )

      Obviously you need to show proof of citizenship to buy the computer

      Why? It's going to be made in China.

  • If they're concerned about national security, they should change the administration.

  • AWS researchers supposedly found the "jailbreak", meanwhile everybody else (MS, FireEYE (Mandiant) aka Google) missed it, seems somewhat plausible

You are in a maze of little twisting passages, all alike.

Working...