Think you can use American products? Think again. In fact, every company in the world right now needs to be moving off American technology or preparing for a move, because Trump just showed how he can shut it down like oil.
Anthropic had released Fable two months after declaring its models too dangerous to be released. Fable was said to be proof that safety was solved, soon after they had said the opposite. What it in fact delivered was a machine that refused to work.
This has become a rather sad theme in the AI industry. Zero integrity control, causing outages.
The “service safety” they designed was a denial of service, the literal opposite. The detection system was so aggressive that ordinary requests were kicked out. A model that shuts down at a benign prompt is not secured. It is broken. That is insecurity, by definition. The company called insecurity their security release. They sold denial of service as a premium feature and labeled protection.
The United States read the Anthropic marketing about dangers and believed it. Hook, line and sinker! The gullible and unpredictable Trump administration demanded access be revoked for a thing that couldn’t reliably be accessed. On 12 June the Commerce Department issued an immediate export control directive suspending all access to Fable 5 and Mythos 5, the first time the federal government has reached into a deployed commercial model and switched it off.
To be accurate, Anthropic had never designed access around denying foreigners. The Trump administration is obsessed with punishing people based on an “alien” designation (“artfremd” in Nazi Germany). So the Trump order was to deny access to foreigners and Anthropic was forced to disable it for everyone instead.
This is reminiscent of 2009-era stuff when the initial cloud providers were shocked, shocked I tell you, to find out ITAR is real. Here we are today with Anthropic claiming they can’t tell a foreign national from anyone else in real time, so the model is down. As if ITAR is real. Again.
Perhaps the right way to read all of this is that the American government just demonstrated a footgun clumsily and loudly, that at any minute they will cut off the world from American-based technology. The Straight of Hormuz disaster is just the beginning. They did it with Microsoft, now they are targeting Anthropic.
The warning shot to the world is the big, centrally distributed American services aren’t trustworthy. They are at the whim of Trump. Just like oil.
The stupid part of the story, aside from America being a hot-headed tin-pot dictatorship under the thumb of ex-KGB, is that Anthropic is pleading that there’s no danger to the public, while the government still believes the Anthropic claims that there’s danger to the public.
…the level of capability displayed there is widely available from other models (including OpenAI’s GPT-5.5), and is used every day… 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.
Anthropic says whoa you can’t just block a model for being told dangerous, only we can do that as a marketing trick.
The capability the government feared, by Anthropic’s own account, is a model reading a codebase and fixing flaws, something it concedes other public models already do without restriction. So the Trump order protects no one from a tool the vendor now insists is ordinary.
A product that shut itself down aggressively to prevent usage has been certified as a weapon and withdrawn from a public that could never use it in the first place. Even the jailbreak is downplayed by Anthropic as nothing to fear.
The Anthropic incoherence is a huge problem, as I have pointed out flaws with their Mythos system card and again flaws with their Fable system card.
For example, look at how an Anthropic card in 3.1.1 opens by calling Mythos 5 the most capable cyber model they have ever evaluated, then in the next breath files it in Tier 1, their lower risk category, defined as human-dependent assistance with known techniques. I can not get over how dumb America looks right now. Incoherence and disproportion take the main stage, not any absence of capability. The Trump ban is for the low tier by definition. Reading a codebase and fixing previously known bugs is textbook Tier 1: meaningful assistance with known methods, still needing a human. Arguably beneath it, since it is defensive. So America has openly stated to the world an export control blocks a Tier 1 interaction over a capability Anthropic reserves for Tier 2.
It’s like Trump banning GPU export because they can generate images. Dude, it’s what people buy them to do. That’s not even dual-use, that’s just use. And then we have to deal with the fact that Anthropic designed a thing that powers off when you power it on.
Unusable once by design.
Unusable twice by decree.
American integrity is twisting in the breeze like an infamous bridge we were all supposed to study so it would never happen again.