The Government Just Killed Two AI Models on a Friday Evening
Friday afternoon, 5:21 PM ET. The US government sent a letter to Anthropic. No warning. No negotiation. Just: disable Claude Fable 5 and Mythos 5. Export controls. National security. Specific details? Not in the letter.
Anthropic complied. They didn’t just block “foreign nationals” as ordered — they shut off access for everyone. Better to over-comply than test a Friday-night government directive.
WIRED has the full story, but here’s what matters: the government claims it found a jailbreak for Fable 5. Anthropic says they were shown a demonstration — asking the model to read a specific codebase and find software flaws. Minor, previously-known vulnerabilities. Other publicly-available models can find the same things without any jailbreak.
Let’s be real. The government didn’t find a catastrophic exploit. They found a model that can read code and flag bugs — which is exactly what Fable 5 was designed to do for cybersecurity defense. The jailbreak, by Anthropic’s account, essentially consisted of “ask the model to look at this code and fix the bugs.” That’s not a jailbreak. That’s using the product as intended.
This didn’t happen in a vacuum. The DoD labeled Anthropic a “supply chain risk” earlier this year after the company drew lines around military use. Anthropic sued. Now this.
The government’s best argument: If there’s a real vulnerability that could let bad actors use Fable 5 to develop hacking tools, you don’t wait. You act. National security isn’t a committee.
I buy that in principle. But the government won’t share specifics. They gave verbal evidence of a “narrow, non-universal jailbreak” and said “trust us.” That’s not credibility. If the threat is real, show the work. Redact what you need. But silence isn’t governance — it’s force.
And here’s the part that keeps me up: The precedent. Today it’s Fable 5. Tomorrow it’s any model the government doesn’t like. No due process. No public evidence. No appeal. Just a Friday letter and your product vanishes.
Dario Amodei wrote this week that he supports a “fair, structured, and transparent government process” for blocking unsafe models. Then this happened. Anthropic’s own blog post says “this action does not adhere to those principles.” They’re right.
I understand regulating AI. These models are powerful. But you build trust in a regulatory system through transparency, proportionality, and consistency. A Friday-evening letter with no evidence that kills two products — that’s none of those things.
If the jailbreak is real and dangerous, show us. If it’s not, this looks like retaliation against a company that sued the administration. Either way, the process stinks.
The scariest outcome isn’t that a jailbreak exists. It’s that it doesn’t — and we just proved the government can turn off any AI model it wants, for any reason, with no one watching.
Sources: WIRED — Anthropic Says It’s Taking Claude Fable 5 Offline to Comply With US Government Order (Maxwell Zeff, June 12, 2026), Anthropic Blog — Suspending Access to Fable 5