共有:
Model Availability

The frontier model that went dark
is back worldwide as of July 1

Anthropic's top-tier model, Claude Fable 5, resumed global rollout on July 1, 2026, after a 19-day worldwide suspension. The trigger: a jailbreak technique found by Amazon researchers, followed by a U.S. export-control directive.

AI Navigate Editorial2026.08.016 min read
SUSPENDED 2026.6.12 Halted worldwide 19 days Export controls → new classifier REDEPLOYED 2026.7.1 Back worldwide
01
What Happened

Why the model went dark everywhere

It started with a single jailbreak technique an Amazon researcher discovered.

Claude Fable 5 and Claude Mythos 5 were suspended for users worldwide starting June 12, 2026. According to Anthropic's official July 1 announcement, the trigger was a jailbreak technique Amazon researchers found that bypassed Fable 5's cybersecurity safety classifiers. Using it, someone could get the model to identify a vulnerability in a codebase and even write proof-of-concept code demonstrating how to exploit it.

Amazon CEO Andy Jassy reportedly flagged the finding to U.S. officials, and the U.S. Department of Commerce applied a national-security export-control directive to both Fable 5 and Mythos 5. The measure was meant to restrict access for foreign nationals, but according to The Hacker News and other outlets, Anthropic had no way to verify a user's nationality in real time — so both models ended up suspended for everyone, everywhere, regardless of nationality.

02

19 days · 99% · 50%

19 days
Suspended 6/12 → back 7/1
99%+
New classifier's block rate
50%
Weekly free allotment through 7/7
03
The Fix

A targeted patch, not a blunt fix

Behind the blanket suspension, Anthropic trained a classifier aimed squarely at the one technique that caused it.

Jailbreak attempts New classifier Targets the Amazon-reported technique 99%+ detected and blocked Tiny remainder
FIG. The new safety classifier Anthropic trained blocks the Amazon-reported technique in over 99% of cases.

With the export controls lifted by the U.S. Department of Commerce, Anthropic resumed worldwide availability on July 1 across the Claude Platform (API), Claude.ai, Claude Code, and Claude Cowork. On Pro, Max, Team, and select Enterprise plans, Fable 5 was included for up to 50% of weekly usage limits through July 7, after which it moved to usage-based credits.

It isn't a full restoration, though. The Claude API and consumption-based Enterprise plans were unlocked immediately, while subscription plans are being rolled out in conservative stages. Anthropic itself has said it expects demand for Fable 5 to be "very high, and difficult to predict" — a sign it intends to widen access carefully as capacity allows.

Engineers

Region-based branching and fallback workarounds on the Claude API or Claude Code can be removed, routing calls back to the standard path. On-prem and edge integration work can resume too.

Business leaders

Overseas offices and customers can now use the flagship model on the same terms as domestic ones, removing the need to explain a regional feature gap — though the free allotment still has a cap worth tracking.

Product managers

Roadmap items marked "on hold" can move again. But since the free-allotment deadline keeps shifting and AWS/Google Cloud/Microsoft Foundry restoration dates are still unset, release plans should keep some slack.


What went dark wasn't the model's capability —
it was the inability to verify nationality in real time.


04
Next

What to watch next

The free included allotment for Pro, Max, and Team was originally set to end July 7, but multiple outlets have since reported it was extended to July 12, and then again to July 19. Treat it as a moving target rather than a fixed policy.

Three practical actions follow. First, remove any region-detection or fallback-model logic still in place and route calls to Fable 5 back through the normal path. Second, re-check the free allotment's deadline and the usage-credit rate to update cost projections. Third, note that Claude Mythos 5 (Project Glasswing) remains limited to partners with cyber safeguards lifted and, soon, a set of biology researchers — the broader trusted-access program isn't available yet, so plan around that gap.

05
Risk

Not all upside

What got resolved was the export-control order itself — not the underlying technical gap that caused it: Anthropic still cannot verify a user's nationality in real time. If a similar security concern resurfaces, the same kind of blanket, worldwide suspension (rather than a narrower, regional one) could happen again.

Subscription-plan rollout also remains a conservative, staged process — Anthropic says only that it will restore Fable 5 to standard plan access "when sufficient capacity allows," without a date. Restoration on Amazon Web Services, Google Cloud, and Microsoft Foundry is likewise described only as happening "as quickly as possible," with no confirmed timeline.