The US government pulled the newest model from Claude-maker Anthropic offline this month on national-security grounds, and the company took it down worldwide rather than risk breaking the order. Now President Trump is hinting at when it might be allowed back.
If you don’t follow artificial intelligence closely, here’s the quick version. Anthropic is the American company behind Claude, one of the main rivals to ChatGPT. On 9 June it released its most capable system yet, in two versions: a consumer model called Fable 5, and a more powerful, fewer-guardrails model called Mythos, meant for vetted users like governments. It was originally called too powerful to release or an AI ‘super weapon’.
Three days later, both were gone.
What happened
On 12 June, the US Commerce Department issued an export-control order. It barred access to the two models by any foreign national, whether inside or outside the United States, and it even covered Anthropic’s own overseas staff.
Export controls are the trade-law powers Washington normally uses to keep sensitive technology, think advanced chips or weapons parts, out of the wrong hands. Turning them on a commercial AI product that was already live to the public is, by most accounts, a first.
Anthropic couldn’t screen every user by nationality in real time, so to comply it switched both models off for everyone, everywhere.
What set it off
The trigger, according to multiple reports, was a warning from Amazon, which is both an Anthropic investor and a competitor in AI. Amazon researchers said they had prompted the model into producing restricted information related to cyberattacks.
From there the two sides disagree sharply, and the disagreement is the whole story.
The administration’s account, laid out publicly by White House AI adviser David Sacks, is that this was a genuine security hole the company was asked to fix or pull, and that it declined. Anthropic says the finding was “a narrow potential jailbreak” that gave no real uplift to a bad actor, and called the shutdown “a misunderstanding”. Several independent security researchers took Anthropic’s side; Katie Moussouris, a well-known figure in the field, said flatly: “It’s not a jailbreak.”
A jailbreak, in this context, means tricking an AI into ignoring its own safety rules and handing over something it’s built to refuse. Whether what Amazon found was a dangerous one or routine security testing is the point still in dispute.
The “part owner that turned them in”
In an interview with Axios on 19 June, Trump pointed back at that warning. “It was a competitor and a part owner that turned Anthropic in,” he said. “They didn’t like what they were doing. They were very concerned.” He didn’t name Amazon. Reporters did though.
That ownership tangle matters for investors, because it’s about the only way most people can get near Anthropic. The company is privately held, reported to have been last valued at around US$965 billion, so there’s no stock to buy. Its big backers are listed, though. Amazon has committed many billions of dollars, Google parent Alphabet owns a stake reported at about 14 per cent, and chipmaker Nvidia has pledged up to US$10 billion. None of them controls the company, which is run as a public-benefit corporation.
When does it come back?
That’s the open question, and Trump’s answer was warmer than it would have been a week earlier. Asked whether he saw Anthropic, and its chief executive Dario Amodei, as a national-security threat, he said: “Not now, but a week ago, maybe.” The two had met at the G7 summit in France days before.
Trump said Anthropic had “behaved very responsibly to our request”, and played down the need to reach for tougher tools. Asked whether he would use the Defense Production Act, a Cold-War-era law that lets a president direct private companies on security grounds, he said: “I would, but I’m not sure I have to do that.”
As of 22 June, the models were still offline and the Commerce order still stood. Anthropic has offered refunds to customers who paid to upgrade, and says it expects access to return “as soon as possible”. No date has been set.
