After 19 days offline, Anthropic’s most powerful AI model is back. The US Commerce Department lifted export controls on Claude Fable 5 and Mythos 5 on June 30, 2026, and Anthropic began restoring global access on July 1 — adding a new safety layer in the process.
What Triggered the Shutdown
On June 12, the US government ordered Anthropic to suspend access to Fable 5 for foreign nationals after an internal Amazon research report showed a specific prompt could bypass certain safeguards and identify software vulnerabilities. Anthropic had no choice but to pull both models offline within hours, disrupting thousands of enterprise customers worldwide.
The move was unprecedented: a US government export-control order forcing a commercial AI company to take a live product offline mid-deployment. It put AI infrastructure risk back on the boardroom agenda and raised hard questions about single-provider dependence at the frontier tier.
What Changed to Get It Back
Commerce Secretary Howard Lutnick confirmed that Anthropic no longer requires an export license after agreeing to three commitments: proactively detect and address security risks in its models; work with the US government on protocols for all future model releases; and report any malicious activity discovered within its systems.
Critically, Anthropic also trained a new safety classifier specifically designed to detect and block the bypass technique described in Amazon’s original report. The company says the updated classifier blocks the reported exploit in more than 99% of cases.
What Is Live Now
As of July 1, Fable 5 is rolling out globally across Claude Platform, Claude.ai, Claude Code, and Claude Cowork. Anthropic will restore July 7 usage credits to affected subscribers and API customers to compensate for the 19-day window.
Mythos 5, the restricted vulnerability-hunting variant, is likewise being restored to its approved partner network under Project Glasswing — Anthropic’s joint cybersecurity initiative covering 150+ organizations across 15 countries.
What This Means Going Forward
The agreement sets a new precedent for how the US government engages with frontier AI labs. Rather than broad export restrictions, the framework now requires labs to proactively collaborate with Commerce on security reviews before releasing their most capable models — a model that other labs will almost certainly be asked to follow.
For enterprises, the 19-day outage surfaced a real infrastructure risk: dependence on a single AI provider at the frontier tier. Expect to see more AI business continuity planning in procurement conversations going forward.
The resolution also clears a cloud over Anthropic’s IPO. The company filed a confidential S-1 at a $965 billion valuation, and the export-control saga had raised questions about government-relationship risk. With a formal cooperation framework now in place, Anthropic enters the IPO window in a stronger regulatory position.
FAQ
Why did the US government restrict Fable 5?
An Amazon research report showed a specific prompt could bypass Fable 5’s safeguards to identify software vulnerabilities. The Commerce Department applied export controls to restrict foreign access while the issue was resolved.
Is Fable 5 safe to use now?
Anthropic says its new cybersecurity classifier blocks the reported exploit in more than 99% of cases. The model is now live globally for all previously approved users.
Does this affect Claude Sonnet 5 or other models?
The export controls applied only to Fable 5 and Mythos 5. Claude Sonnet 5 and other Claude models remained available throughout the 19-day window.
