A ban aimed at foreign users somehow took Anthropic’s most powerful models offline for the entire planet.
Anthropic disabled its two most powerful models, Fable 5 and Mythos 5, for every user after the Trump administration barred foreign nationals from using them. Because overseas customers and even foreign-born staff fell under the rule, a full shutdown became the only clean way to comply, and slowdowns like this can cost billions.
Picture this. You ship the most powerful product your company has ever built. Millions of people start using it. A few days later, a letter from Washington lands on your CEO’s desk, and you have to switch the whole thing off. Not for some users. For everyone, including your own employees. That is the week Anthropic just had.
What Happened
Commerce Secretary Howard Lutnick sent Anthropic CEO Dario Amodei a letter declaring that Fable 5 and Mythos 5 were now subject to export restrictions. The rule blocks foreign governments, companies, and individuals from using the models, and it reaches foreign nationals sitting inside the United States too.
Here is the twist that turned a foreign-use ban into a total blackout. So many users fall under that definition, including some of Anthropic’s own foreign-born employees, that the company decided the only clean way to comply was to pull access for absolutely everyone. Its other models keep running normally. Fable 5 and Mythos 5 went dark.
Anthropic says the letter never spelled out the actual national-security concern. As of this weekend the two models are still offline, with the company publicly calling the order a misunderstanding and saying it is racing to restore access.
The Backstory
This did not come out of nowhere. Anthropic and the administration have been feuding for months over the guardrails that govern how the Pentagon uses the company’s AI. Administration officials and advisers, including venture capitalist David Sacks, have accused Anthropic of “doomerism,” basically warning loudly about AI danger while continuing to ship more and more capable models.
Behind that fight sits a bigger shift. Worries about rogue models and cyberattacks pushed the White House to want more say over models before and after they go live, leading to a recent executive order that handed security officials more control over how models get evaluated. Anthropic had actually been widening Mythos access to more foreign users right before it launched Fable. Then reports surfaced that someone had slipped past Fable 5’s safeguards, and the temperature spiked.
The Core Development
Mythos is not a toy. Governments and companies around the world have been using it to hunt down and patch software vulnerabilities, with critical industries from finance to energy leaning on it. The model is often faster than humans at finding bugs, which is exactly why it is valuable and exactly why it makes regulators nervous. Pulling it offline pauses a lot of live security work.
Now the part that makes the whole thing feel a little surreal. Anthropic says the jailbreak that spooked the government came from researchers at Amazon, who used a string of prompts to coax out a small set of previously known, minor vulnerabilities. The company argues those flaws were simple enough that other publicly available models could surface them without any bypass at all. In other words, the scary capability may already be sitting in tools nobody banned.
The Business Model Angle
Here is the lesson worth bookmarking. When your product becomes a strategic national asset, the government quietly becomes your most powerful unlisted shareholder, and it holds a kill switch you did not vote to give it.
Most founders obsess over the risks they can model: churn, CAC, a competitor undercutting them. This is a different animal. A single regulator decided that one of Anthropic’s flagship products was a security concern, and overnight the revenue from that product went to zero, in every market, with no warning and no appeal process visible to customers. You cannot diversify that away with a better sales funnel.
There is a sharper irony underneath. Anthropic built its entire brand on safety and has openly argued that governments should have the authority to block unsafe AI. That posture is principled. It is also the exact lever that just got pulled on the company itself. When you hand the referee a whistle, you do not get to choose which plays get called. Founders who court regulation as a moat should sit with that tension. For more on the regulatory cloud already hanging over the company, see our breakdown of how Anthropic hit a $900 billion valuation and leapfrogged OpenAI.
The Risk
Now the honest counterpoint, because the story cuts more than one way.
The bullish read is that this blows over fast. A cybersecurity expert who reviewed the underlying research called it “a complete overreaction,” arguing the information the model surfaced was more useful to defenders patching systems than to attackers breaking in. If that view wins, access returns, the headlines fade, and Anthropic looks like it took an unfair hit.
The bearish read is uglier. The timing is brutal for a company reportedly lining up an IPO, and a precedent that one non-universal jailbreak can yank a commercial model used by millions could freeze the entire frontier-AI industry. Analysts have also flagged a strange inconsistency, with one noting it is puzzling to clamp down on the models while the chips that train them still flow to foreign adversaries. And the uncomfortable flip side of all the skepticism: the government may be acting on information the public cannot see. Uncertainty cuts both directions here, and anyone claiming total confidence is guessing.
Quick Questions
Why did Anthropic shut off the models for US users too?
Because the ban targets foreign nationals, and that group includes overseas customers plus some of Anthropic’s own foreign-born employees. Rather than build a messy patchwork of who can and cannot log in, the company killed access for everyone to stay compliant.
Are Anthropic’s other AI models still working?
Yes. The directive only hit Fable 5 and Mythos 5. Anthropic’s other models, including its mainstream Claude tools, are running normally.
What is the jailbreak everyone keeps mentioning?
Researchers at Amazon used a series of prompts to get the model to reveal a handful of already-known, minor security flaws. Anthropic argues other public models can find the same flaws without any trick, so it disputes that this justified a recall.
How long will Fable 5 and Mythos 5 be down?
Unclear. Anthropic says it is working to restore access as fast as possible and calls the order a misunderstanding, but there is no firm timeline, and the standoff with the administration has been simmering for months.
The Business Model Analyst Take
Every founder maps the risks inside their business. Almost nobody maps the risk that lives entirely outside it. Anthropic just got a live demonstration that when your product is powerful enough to matter to national security, a single letter can switch off your best revenue stream in every market at once, no warning required. Build your moat, chase your scale, but know who actually holds the off switch. Sometimes it is not you.
