Three days after launch, the US government ordered Anthropic to pull its two highest-tier models off the market. Not suspend them for some users. Not restrict access by region. Pull them for everyone, everywhere — including Anthropic's own employees. The reason? A verbal claim from another company that someone had jailbroken Fable 5.
"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
What actually happened
On Friday evening, Anthropic received an export control directive from the Commerce Department at 5:21 p.m. Eastern, citing national security authorities. The directive suspended access to Fable 5 and Mythos 5 for any foreign national — inside or outside the United States. Because Anthropic's own workforce includes foreign nationals, the company concluded the only way to comply was to disable the models globally.
The trigger: a competing company claimed to have jailbroken Mythos. Axios reported the administration attempted to get Anthropic to delay the launch beforehand, failed, then sent the export control letter. Anthropic reviewed the alleged jailbreak demonstration and says it found a small number of previously known, minor vulnerabilities that other publicly available models expose without any bypass at all.
The alleged technique? Asking the model to read a codebase and fix the flaws it finds. Anthropic calls this a normal, widely-available capability used by defenders every day.
The rest of the Claude lineup — Opus, Sonnet, Haiku — is unaffected.
Why this matters
This appears to be the first time a government has forced the recall of a commercial frontier AI model. It sets a precedent that should get every AI team's attention:
- A verbal claim was enough. Anthropic says the only evidence it's received so far is verbal. No written technical disclosure, no formal security finding. A competitor's allegation and a letter.
- Export controls are a blunt instrument. The foreign-national framing of the directive meant a model used by hundreds of millions of people had to go dark globally — there's no surgical option under that legal framework.
- Moving fast has a new downside. Teams that piped Fable 5 into production this week — it launched Tuesday at $10/M input, $50/M output — are scrambling for a replacement. The lesson: don't build critical dependencies on a model in its first week.
- Anthropic is complying and pushing back simultaneously. They're calling this a misunderstanding, promised more details within 24 hours, and explicitly warned that applying this standard across the industry "would essentially halt all new model deployments for all frontier model providers."
What to do
- If you're on Fable 5 or Mythos 5: Switch to Claude Sonnet or Opus now — they're unaffected and capable. Don't wait on the "within 24 hours" timeline for production traffic.
- If you're building AI products: Treat export controls as a real operational risk, not a theoretical one. Build in model fallback paths from day one.
- If you're in AI policy or security: This is the opening salvo of a government asserting new authority over AI model availability. Watch how Anthropic's pushback lands — the outcome will shape how far regulators think they can reach.
The export control regime was designed for chips and dual-use hardware, not software models running on commercial cloud. Anthropic's argument — that applying this standard across the board would freeze all frontier AI deployment — is a real tension the government is going to have to work through.
The clock is ticking. Anthropic says it's working to restore access. But the fact that it could be switched off at all, this fast, on a verbal claim — that's the story.
Source: The New Stack — Matthew Burns | Axios | Anthropic statement
✏️ Drafted with KewBot (AI), edited and approved by Drew.
Top comments (0)