Days after opening Fable 5 to the public, a US government order forced Anthropic to pull it

Anthropic disabled Claude Fable 5 and Mythos 5 on Friday after a US government order. The two models went dark for every user worldwide, days after Fable 5 became the company’s first Mythos-class model open to the public. Last week Anthropic marketed Fable 5’s coding gains; this week it can’t serve the model at all.

The instrument is an export control directive from the Commerce Department, signed by Secretary Howard Lutnick. It bars access to both models “by any foreign national,” a category Anthropic reads to include non-citizen employees inside its own offices. Honoring that line item meant pulling the models entirely rather than fencing off a region. This looks like the first time a leading AI lab has taken a publicly deployed model offline because the federal government told it to.

What we know

The shutdown was fast and total. Anthropic received the directive at 5:21 PM ET on June 12 and says it disabled the models the same day, citing “national security authorities.” Here’s the confirmed picture.

• Simon Willison watched the API go dark in real time, logging a final successful Fable 5 call before a 404 took over. “Well this is nuts,” he wrote.
• It hit downstream platforms fast. Vercel posted that “in compliance with the US Government’s directive, access to Claude Fable 5 is now suspended on AI Gateway for all users,” and that it doesn’t “know if or when access to the model will be restored,” per its changelog.
• The trigger was a reported jailbreak. The government believed it had a method to bypass Fable 5’s safeguards, 9to5Mac reported, and treated the cybersecurity capability as an export risk.
• Anthropic pushed back on the severity. Its statement says the issue was “relatively simple” and “already available in other publicly-available models, including OpenAI’s GPT-5.5.” That capability, the same one we covered when Mythos 5 went to US cyberdefenders, is “used every day by the defenders who keep systems safe.”
• Everything else still works. Opus 4.8, Sonnet 4.6, Haiku 4.5, and earlier models keep running on Anthropic’s API, AWS Bedrock, Google Vertex, and partner gateways. Tool-calling over MCP is unaffected for those models.

What we don’t know

For all the speed of the takedown, the path back is undefined. As of June 13 there’s no public timeline and no technical paper backing the government’s claim.

• No restore date. Anthropic says only that it’s “working to restore access as soon as possible,” with no timeline and no clear path back through licensing.
• The full technical claim is murky. Willison notes the government’s evidence for the jailbreak was described verbally as “a potential narrow, non-universal” bypass, not a published proof.
• Who blinks first. Whether this resolves through a granted export license, a narrowed directive, or a longer standoff with the Bureau of Industry and Security is open.

Where the reporting comes from

Anthropic’s own statement is the primary record. The directive’s authorship, Lutnick and the Bureau of Industry and Security, was reported by CNBC and 9to5Mac. Vercel’s changelog confirmed the gateway-level cutoff. This follows a pattern: the administration had already signaled a tighter grip in the narrower AI executive order earlier this year.

What this means for you

If you wired anything to Fable 5 this week, it’s already throwing 404s, and there’s no ETA. Treat the model as gone for now and fail over to Opus 4.8 or Sonnet 4.6, both of which are untouched on every gateway. Pin model IDs explicitly so a silent fallback doesn’t reroute traffic you didn’t budget for. The bigger lesson sits above the API: a frontier model can now vanish on a Friday afternoon because of an export order, not an outage or a price change. If your product depends on a single top-end model, that’s a supply risk you can’t patch in code. Keep a second vendor’s model qualified and ready, and don’t build a flagship feature on a model you can’t serve to half your users.