Anthropic Restores Claude Fable 5 Globally After US Export Controls Lifted

For anyone who uses Claude for work, study, or everyday tasks, the past three weeks have been frustrating. On 12 June 2026, just days after Anthropic launched its powerful new Claude Fable 5 model, the United States government imposed export controls on it — and on the related Claude Mythos 5 — forcing Anthropic to pull global access almost immediately after launch. Now, following talks between Anthropic and US officials, those controls on Fable 5 have been lifted. The model returns to worldwide users from 1 July 2026.

It’s a sharp illustration of just how seriously governments are starting to treat frontier AI. One week you can access a state-of-the-art reasoning model; the next, a government order takes it offline.

What Is Claude Fable 5?

Claude Fable 5 is Anthropic’s latest publicly available large language model, sitting within what the company calls its “Mythos-class” generation. Designed for complex reasoning, software engineering, knowledge work and vision tasks, it was made generally available on 9 June 2026 through the Claude API, Claude Platform, Claude Code and Claude Cowork, as well as via AWS, Google Cloud and Microsoft Foundry.

Think of it as the version of Anthropic’s most capable model that’s been made suitable for broad release — the company has layered in safety guardrails that its more restricted sibling, Claude Mythos 5, does not carry for general users. Mythos 5 itself remains limited to approved US organisations through a scheme called Project Glasswing.

Fable 5 is priced at around $10 (about £8) per million input tokens and $50 (about £40) per million output tokens via the API — roughly double the token price of Claude Opus 4.8. Subscription plan users on Pro, Max and Team tiers were given promotional access windows where Fable 5 usage counted against their weekly usage limits, with further use requiring pay-as-you-go credits.

Why Did the US Government Step In?

Export controls are legal tools that allow governments to restrict the transfer of certain technologies to overseas users, typically on national security or foreign policy grounds. US authorities grew concerned about the potential for highly capable AI models like Fable 5 and Mythos 5 to be misused for offensive cybersecurity purposes — things like system exploitation, malware creation, or attacks on critical infrastructure.

The controls came into force on 12 June 2026, just three days after launch. Anthropic suspended global access to both models.

But the suspension didn’t last. Following what Anthropic described as “productive conversations” with US officials, the export-control order on Claude Fable 5 was lifted on 30 June 2026. The conditions: Anthropic would introduce a new layer of safety classifiers specifically tuned to detect and block cybersecurity-related prompts judged to be high-risk.

Stronger Cybersecurity Guardrails — With Trade-Offs

The new classifiers are designed to aggressively detect prompts related to offensive cybersecurity — things that could help someone find vulnerabilities, build malware, or gain unauthorised access to systems. When Fable 5 encounters such prompts, it will refuse, or in some cases fall back to Claude Opus 4.8 to handle the request more cautiously.

Anthropic is open about the downside. In the near term, the company acknowledges, the tighter classifiers may accidentally catch legitimate requests — routine coding tasks, defensive security work, or cybersecurity-adjacent research that isn’t harmful at all. The company says it will refine the system over time.

Not everyone is happy about that. Some developers and security researchers have already pushed back, arguing the classifiers are too blunt. Civil liberties groups have raised broader concerns about export controls concentrating power over advanced AI in a small number of companies and governments. And business users — many of whom built workflows around Fable 5 in the brief window before the suspension — have pointed out that sudden outages like this make it risky to depend on frontier models for anything critical.

Those are fair concerns. And they’re unlikely to go away as AI models become more capable and governments become more interventionist.

Anthropic’s Position

Anthropic has been careful to frame its cooperation with US authorities as part of a broader commitment to what it calls “frontier AI safety.” The company says it is working with governments and regulators on a “trusted access” framework for its most capable models, which includes export-control compliance and domain-specific safeguards.

Dario Amodei, Anthropic’s chief executive, has previously argued that safety and capability are not in conflict — that it is possible to build powerful AI and deploy it responsibly. The Fable 5 episode puts that argument to a practical test.

Redeployment from 1 July 2026 covers Claude.ai, the Claude Platform, Claude Code and Claude Cowork. Access via AWS, Google Cloud and Microsoft Foundry should be re-enabled as soon as possible, though Anthropic has not given a precise date for those channels.

What This Means for Kent Residents

Anyone here in Kent using Claude.ai or Anthropic-powered tools — whether for work, study, or personal projects — will regain access to Fable 5’s higher-end capabilities from 1 July 2026, provided their subscription or service plan includes the model. Developers, IT teams and cybersecurity professionals in the county should be aware that the redeployed model carries tighter restrictions on security-related prompts, which may occasionally interrupt legitimate coding or testing workflows and require switching to an alternative model. Students and researchers at the University of Kent and Canterbury Christ Church University working in computer science or cybersecurity may also need to factor the new guardrails into any coursework or projects that rely on Claude Fable 5. More broadly, the episode is a reminder for any UK business or public-sector body building on frontier AI that government intervention in model availability is now a real operational risk to plan for.