U.S.-based AI company Anthropic recently released two AI models, Fable 5 and Mythos 5. However, the U.S. government reportedly raised national security concerns over these advanced AI systems. As a result, the government directed the company to restrict access to the models and prevent foreign nationals from using them.
The company explained the entire situation in a statement published on its website, detailing what happened in the case and the reasons behind the government's order.
Anthropic had already been on the Trump administration's radar even before this incident.
In April, Anthropic did something strange. They announced an AI model — and immediately said they weren't releasing it.
The model was Claude Mythos. Too good at finding software vulnerabilities. The kind of good that makes cybersecurity experts nervous.
So instead of a public launch, Anthropic handed it to approximately 150 vetted organizations—security firms, infrastructure providers, and researchers. Use it to find flaws before the bad guys do.
The results? Over 10,000 critical vulnerabilities have been found in real systems. That's either reassuring or it isn't, depending on how you sleep at night.
"It's the unknown unknown." —Canadian Finance Minister François-Philippe Champagne, on why Mythos had everyone's attention.
So here's what Anthropic did. And look, it's kind of genius, kind of not. Take Mythos. Keep the brain. Add a speed limiter. Call it Claude Fable 5.
Same underlying model. Same training. Just with guardrails blocking the dangerous stuff—cybersecurity queries, biology, and chemistry. High-risk requests get quietly rerouted to an older, safer model.
The benchmarks were hard to ignore. On SWE-Bench Pro — real-world software engineering tasks from actual GitHub repos — Fable 5 scored 80.3%. The previous best? 69.2%.
The longer the task, the bigger the gap.
Anthropic didn't pretend there was no risk. "Releasing a model this capable comes with risks."
They wrote it—and released it anyway.
An IPO at a near-trillion-dollar valuation tends to sharpen the timeline a little. Which is a number that stopped meaning anything to me around the time it hit $200B, but here we are.
Apr 7
Mythos was revealed privately. Too dangerous for the public. Handed to 150 orgs.
Jun 9
Fable 5 goes public — same model, guardrails added. Live on APIs, AWS, Azure, and Google Cloud.
Jun 12
Government steps in—export control directive issued at 5:21 PM ET. No access for any foreign national, anywhere.
Jun 13
Global blackout — Anthropic disables both models for everyone. Sessions return errors. Older models step in.
The US government's order was surprisingly sweeping. No access for any foreign national — including Anthropic's own non-citizen employees.
To comply with that literally, you have to shut it down for everyone. Which is exactly what happened.
The reason? Someone apparently found a way to jailbreak Fable 5. To get past the guardrails. To access the Mythos capabilities underneath.
The government found out. Moved fast. Didn't explain much.
"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."
— Anthropic, official statement, June
Anthropic's point isn't wrong. Every model gets jailbroken eventually. The real question is, how bad is the harm window, and how fast can you close it?
And I'm still not sure what to make of the jailbreak part, honestly. Like, was it serious? Was it a demo exploit someone posted on Reddit? Nobody's saying anything.
Regulators, apparently, weren't in the mood to debate that.
Anthropic co-founder Jack Clark said something last week that's hard to shake. The AI industry has a gas pedal, he told the BBC. No brake pedal.
"You want the option to be able to take your foot off the gas and put your foot on the brake," he said. "Right now, that option doesn't exist."
Cool metaphor. Also, he works at Anthropic. Just saying
The Fable 5 story is that tension made visible. Guardrails that looked solid on Tuesday had gaps by Friday. A government order arrived without explanation. A trillion-dollar company had to go dark in six days.
What happens next? Anthropic says they're working on restoring access. They think it's a misunderstanding.
Maybe it is. Anthropic says they're working on restoring access. They think it's a misunderstanding. Maybe. I don't know. We'll see.
