Anthropic restored Claude Fable 5 to all users on July 1 and gave paying customers a one-week clock: usage counts against half their weekly limit only until July 7, after which every Fable 5 request draws from separate usage credits. Mythos 5, the more capable sibling model, came back only for a set of US organizations approved by the government on June 26. Anthropic said it is still negotiating with federal officials to widen that list.
The restrictions trace back to June 12, when the US government applied export controls to both models after Amazon researchers found a way to prompt Fable 5 into identifying software vulnerabilities and, in one case, generating exploit code. Anthropic pulled access for every user rather than build a real-time nationality check. The export controls lifted on June 30, and Fable 5 returned worldwide the next day on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. Access via AWS, Google Cloud, and Microsoft Foundry is still being re-enabled.
The July 7 date is the detail that separates this from Anthropic’s initial redeployment announcement. It signals the company expects a short grace period rather than an open-ended relaxation of Fable 5’s usage terms, and it puts a concrete cost on the incident for Pro, Max, Team, and select Enterprise customers who relied on the model during the two-week suspension.
Anthropic’s own testing found the vulnerability-identification behavior was not unique to Fable 5. Claude Opus 4.8, GPT-5.5, and Kimi K2.7 could all surface the same vulnerabilities, and models ranging from Claude Haiku 4.5 to GPT-5.4 could reproduce the single exploit demonstration. That finding did most of the work in the government’s decision to lift the controls: the flagged behavior reflected a borderline safeguard trigger, not a Mythos-level capability leak.
Anthropic’s response was a new safety classifier tuned specifically to the technique described in the Amazon report, which the company says blocks that path in more than 99% of attempts. Researchers at the Commerce Department’s Center for AI Standards and Innovation reviewed both the old and new safeguards and endorsed the update. The tradeoff is more false positives on ordinary coding and debugging requests, a cost Anthropic says it will keep tuning down.
Mythos 5 access remains the harder problem. The model was built for defensive cybersecurity work inside Project Glasswing, a group that already includes Amazon, Microsoft, and Google, and its safeguards are lighter than Fable 5’s by design. Anthropic says it is coordinating with the US government to extend Mythos access to more domestic and international partners in that same program, but named no new members and gave no timeline.
Alongside the access restoration, Anthropic is proposing a four-factor severity scale for jailbreaks, developed with Amazon, Microsoft, and Google: capability gain, breadth of capability gain, ease of weaponization, and discoverability. The company also committed to giving designated government partners early access to future frontier models and their safeguards before broad release, plus faster sharing of jailbreak intelligence through the interagency clearinghouse created by the June 2 executive order. Anthropic reported the update on its own site on July 1.
Enterprise teams building on Fable 5 have five days before the usage-credit meter starts. Budget for the switch now rather than discovering the cost mid-sprint, and treat Mythos 5 as still effectively unavailable outside the named Glasswing cohort until Anthropic names new approved partners.
Reported by Anthropic on July 1, 2026.
