Anthropic Restores Claude Fable 5 Globally After 18-Day Export-Control Suspension
TL;DR
Anthropic restored Claude Fable 5 globally on July 1, 2026, ending an 18-day suspension that began June 12 under a US export-control order (Mythos 5 was restored only for a set of US organizations). The order followed an Amazon-discovered jailbreak that made Fable 5 identify software vulnerabilities. Anthropic says a new safety classifier now blocks the technique in over 99% of cases, with blocked requests falling back to Opus 4.8. Fable 5 is live again on Claude.ai, the Claude Platform, Claude Code, and Claude Cowork — but not yet on AWS, Google Cloud, or Microsoft Foundry.
18 days
Length of the global suspension — disabled June 12, restored July 1, 2026
>99%
Anthropic's own claimed block rate for the specific jailbreak technique, per its new safety classifier — a vendor claim, not independently audited
4 surfaces live
Claude.ai, Claude Platform, Claude Code, and Claude Cowork — restored as of July 1
July 7, 2026
Deadline for the 50%-of-weekly-limit included usage on Pro/Max/Team/select Enterprise plans before usage credits are required
Anthropic restored Claude Fable 5 worldwide on July 1, 2026, ending an 18-day global suspension. Mythos 5, however, was restored only for a set of US organizations, following the US government's approval on June 26 — it is not back globally. The arc is now three events in one month: Fable 5 launched June 9, was disabled June 12 under a US export-control directive, and was restored July 1. This article covers only the restoration — for the launch pricing and the disablement rationale, see our earlier coverage.
What triggered the suspension, in Anthropic's account. Anthropic said the June 12 shutdown followed a finding by Amazon researchers that a jailbreak technique could make Fable 5 identify software vulnerabilities in code — and, in at least one reported case, produce working exploit code. The US government treated this as an export-control matter and ordered access suspended for foreign nationals; because Anthropic could not reliably separate foreign nationals from its broader user base, it disabled both models for everyone.
The fix: a new safety classifier, with a stated block rate. Anthropic says it trained an improved safety classifier specifically to catch the jailbreak technique that triggered the suspension. Anthropic states the classifier blocks the technique in over 99% of cases — this is Anthropic's own claim, not an independently verified figure, and operators should treat it as a vendor-reported benchmark rather than an audited result. When the classifier blocks a request, it falls back to Claude Opus 4.8, the same fallback pattern Fable 5 used at launch for its original hard-blocked domains.
Where it's live — and where it isn't yet. As of July 1, Fable 5 is restored on Claude.ai, the Claude Platform, Claude Code, and Claude Cowork. Cloud-platform availability is explicitly "to follow": Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry are not yet back online with Fable 5 as of this restoration. Teams running Fable 5 through any of those three cloud integrations should not assume access has returned — check your specific platform's status before resuming production traffic there.
A short included-usage window, then credits. For Pro, Max, Team, and select Enterprise plans, Fable 5 is included for up to 50% of weekly usage limits through July 7, 2026. After that date, usage beyond what's included requires usage credits, the same mechanism flagged at the original June 9 launch. This is a time-bound allowance, not a permanent entitlement — plan credit provisioning before July 7 rather than after.
What this means for evaluators and Claude Code users. Frontier-model evaluators now have a Fable 5 that has been through one real-world jailbreak-and-patch cycle — worth factoring into any capability-versus-safety comparison against Opus 4.8. Claude Code users who had switched agentic workflows to claude-fable-5 before the suspension can resume on the first-party surfaces listed above; those still routing through Bedrock, Vertex AI, or Foundry remain on Opus 4.8 or another model until Anthropic confirms cloud-platform restoration separately.
Why It Matters
Fable 5 is back, but not everywhere — AWS, Google Cloud, and Microsoft Foundry are explicitly not yet restored, so cloud-platform teams cannot assume parity with Claude.ai and Claude Code. The fix is a classifier update, not a model retrain — Anthropic's >99% block-rate claim is self-reported and worth independent scrutiny before treating the jailbreak risk as closed. The included-usage window is short — the 50%-of-weekly-limit allowance expires July 7, after which usage credits apply, so subscription-plan teams have a narrow runway to plan for the cost shift.
Who's Affected
- — Teams that reverted to Opus 4.8 during the suspension. Fable 5 is available again on Claude.ai, the Claude Platform, Claude Code, and Claude Cowork — you can resume routing to claude-fable-5 on those surfaces, but re-test workflows rather than assuming behavior is identical to pre-suspension.
- — Teams on Amazon Bedrock, Google Vertex AI, or Microsoft Foundry. Cloud-platform availability is 'to follow' and was not restored alongside the first-party surfaces on July 1. Do not assume Fable 5 access has returned on these platforms — check platform-specific status before resuming production traffic.
- — Pro, Max, Team, and select Enterprise subscribers. Fable 5 usage is included for up to 50% of weekly limits only through July 7, 2026. After that, usage beyond the included allowance requires usage credits — provision them before the deadline if you plan to keep using Fable 5 at volume.
- — Frontier-model evaluators and security-conscious buyers. The restoration follows a real jailbreak that reportedly produced exploit code in at least one case. Anthropic's >99% classifier block-rate claim is a vendor figure worth weighing against your own risk tolerance, particularly for workflows touching code analysis or vulnerability research.
What To Do Now
- 1. Confirm which surface you're on before resuming Fable 5 traffic. Claude.ai, Claude Platform, Claude Code, and Claude Cowork are restored; Bedrock, Vertex AI, and Foundry are not. Check your specific integration rather than assuming a blanket restoration.
- 2. Provision usage credits ahead of July 7. The 50%-of-weekly-limit included allowance on Pro/Max/Team/select Enterprise plans is time-bound. If your workload runs at volume, set up credits now instead of waiting for the cutoff.
- 3. Re-test workflows that touch code analysis or vulnerability-adjacent tasks. The new safety classifier targets exactly that jailbreak pattern — expect some previously-allowed prompts to now trigger the Opus 4.8 fallback, and validate that fallback behavior doesn't silently degrade your pipeline.
- 4. Don't treat the >99% block-rate figure as independently audited. It's Anthropic's own claim about its own classifier. Factor that into any compliance or security documentation rather than citing it as a third-party-verified number.
More on this topic — Best AI Coding Tools
Independent Review
Claude AI
Pricing, pros and cons, real-world verdict — no affiliate spin.
Read the Claude AI reviewMore from ToolNav News