Anthropic's most powerful publicly accessible model, Fable 5, is back on Claude after the Commerce Department reversed its June 12 order. The reinstatement is tiered, time limited, and still leaves the cloud and most of the world out.
Three weeks after the US Department of Commerce used an export-control order to pull one of the most capable AI models in commercial use off the internet, the lever has been released. Fable 5, Anthropic's most powerful model that had reached the public, is back on Claude.ai, the Claude API, and Claude Code as of Wednesday, after the Commerce Department's June 26 approval and Anthropic's redeployment announcement on June 30. What arrived is not a full reset. The version of "restored" on offer is narrower, time-limited, and still missing from the cloud routes most enterprise customers buy through.
The original order, issued June 12 by Commerce's Bureau of Industry and Security, removed Fable 5 and the more powerful Mythos 5 from Anthropic's commercial surfaces. Anthropic confirmed the action the same day but disclosed few specifics about the underlying finding. The episode is the first worked example of a US agency pointing an export-control order at a deployed frontier model, and it is also the first time the lever has been released in a way that leaves a partial restoration rather than a return to the pre-order state.
The mechanism is what makes the precedent portable. The US government did not use a compute kill switch, the kind of pre-deployment seizure of model weights that the AI safety community has argued over for two years. It used an export-control order, a regulatory tool normally aimed at chips, chemicals, and dual-use hardware, and pointed it at a running commercial AI system. A Lawfare analysis published after the order was issued argued this is a more readily usable lever than a true compute-side seizure: export controls move through an existing interagency process and require only a national-security finding from Commerce, while a model-weights seizure is far more legally novel. A Center for Strategic and International Studies analysis a week before the lift reached the same conclusion from a policy angle: the June 12 order established a template other agencies can use against other frontier providers.
The access map that emerged on the other side is where the template becomes concrete. Pro, Max, Team, and a subset of Enterprise customers can use Fable 5 inside up to 50% of their weekly usage limits through July 7. After that date, Anthropic shifts Fable 5 from a flat allowance onto metered usage credits, a pricing change the company has not yet publicly detailed. The cap is the visible scar of the three-week gap: even with the ban lifted, paying customers cannot push the model back to its prior weekly consumption level until the credit system is in place. The June 30 lift reporting confirmed the tiered structure but not the post-July 7 credit rate.
The most powerful model in Anthropic's lineup never came back to the public. Mythos 5 was not released to general users even before the June 12 order, and the reinstatement that began this week is limited to Anthropic's Glasswing program, which admits a small set of US organizations pre-cleared with Commerce. International rollout is deferred pending further regulator coordination. The two-model structure is now formally two-track: Fable 5 in a tiered and time-limited form for paying Claude users, Mythos 5 in a US-only, organization-by-organization lane for an unnamed set of cleared customers.
The other missing lane is the one most enterprise buyers actually use. Fable 5 is not returning immediately to AWS, Google Cloud, or other cloud reseller routes, leaving direct Anthropic surfaces as the only paths in for now. Anthropic has told cloud partners that resale is expected to resume "as soon as possible," but no date has been set. A Wired report on the underlying June 26 letter said Amazon's own researchers identified a vulnerability that helped trigger the original finding. If that detail holds, the cloud delay may not be a regulator-imposed hold at all. It may be a separate security review still in progress on the reseller side.
Three questions sit on the watch list. The first is the July 7 cliff: the language in Anthropic's redeployment notice describes a shift from a flat usage allowance to usage credits without publishing the credit price, so the practical cost of Fable 5 access after that date is undefined. The second is the cloud return: the gap between the June 26 Commerce approval and a working AWS, Google Cloud, and Microsoft Foundry listing is the first live test of how cleanly an export-control clearance translates into commercial resale. The third is the precedent itself. Fable 5's return does not unwind the template the June 12 order established. It shows what a restored frontier model looks like after a kill switch has been used on it: present, but on different terms, with the cloud out and the most powerful version still on a leash.