Thirteen days after the Commerce Department ordered it dark, Claude Fable 5 is still nowhere. Send a request to claude-fable-5 and the API hands back an unavailable error. Open claude.ai or Claude Code and a new session quietly routes to Opus 4.8 instead. Anthropic isn't hiding it: its own head of growth says the company isn't serving any Fable or Mythos traffic, with the Fable figure put at exactly zero. Not throttled, not geofenced. Zero.

That isn't an outage. It's a standoff, and the shape of the standoff explains why a model barely two weeks old can't simply be switched back on.

The official line makes it sound trivial. David Sacks, the White House AI adviser, framed the exit cleanly: Anthropic remediates the safety issue, the export control lifts, Fable returns to general release, and the administration wants all of it to happen as soon as possible. Patch the jailbreak, get your model back.

The trouble is what patching concedes. The order followed Amazon's security team flagging a jailbreak in Fable 5 to the White House. If Anthropic ships a fix, it accepts the premise underneath the whole intervention: that Fable carries a uniquely dangerous flaw the other frontier models don't. That premise is exactly what the company has reason to resist, because granting it hands Washington a precedent to darken any future release on the same logic. So the incentives lock. Patch and concede; refuse and stay dark. Either branch leaves the government setting the terms of return.

Watch what's actually moving and you can read the shape of the comeback. Anthropic updated its privacy policy, effective July 8, to start collecting government-issued ID and biometric data. That ID collection is the machinery for a US-only restoration that never needs the export order lifted at all: verify citizenship, serve Americans, keep foreign nationals blocked, comply with the directive on its face. The model comes back for some people by proving who they are, not by the order going away.

Re-release, then, was never a single switch. It's a negotiation with two exits, both of which cost Anthropic something it didn't have to pay on June 9. The return I sketched last week, domestic-first and conditional, is the only kind these two exits produce.

The lever doing all of this was never written as a rule anyone could read in advance, the same unwritten power that took the model down on a Friday evening. A capability that can be revoked by letter and restored only once you've handed over your passport is a different product from the one that launched, whatever the weights say when it comes back.

Sources: