The short version: Anthropic launched Fable 5 on 9 June 2026. The US government imposed export controls on 12 June after an Amazon research report showed a prompt could bypass safety guardrails and identify software vulnerabilities. Non-US users — including all UK businesses — lost access. The controls were lifted on 30 June. Fable 5 returned globally from 1 July, with usage at 50% of normal limits through 7 July while Anthropic phased in the restored capacity.
What the export-control episode actually was
This was not a product recall and it was not routine maintenance. The US Department of Commerce imposed export controls on Fable 5 and the more powerful Mythos 5 under national security regulations — specifically, the Export Administration Regulations that govern technology with potential military or intelligence applications. The trigger was evidence that the model could be prompted to identify exploitable software vulnerabilities more reliably than previous AI models.
Anthropic worked with Amazon, Microsoft, Google, and other Project Glasswing cybersecurity partners to address the specific vulnerability pattern identified before the government approved restored access. The company has also committed to a new joint framework — involving the same partners plus US government agencies — to evaluate jailbreak techniques in future models before public release.
Project Glasswing is Anthropic's programme for trusted cybersecurity partners — organisations with verified security credentials who receive early access to frontier models and participate in safety evaluations. Mythos 5, the more powerful sibling of Fable 5, was cleared earlier for a limited set of US Glasswing partners. Fable 5 is now the general-availability version. Most businesses will work with Fable 5.
What UK businesses can access from 1 July
Fable 5 is available across the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. Pro, Max, Team, and select Enterprise users received access immediately, though at 50% of normal usage limits through 7 July 2026. After 7 July, normal usage limits apply under the existing usage-based credits model.
If your business is on a Free plan, access to Fable 5 depends on plan rollout timing — check your claude.ai interface for model availability. API access to Fable 5 requires an Enterprise tier or specific partner arrangement.
The bigger signal: AI safety is now a government boundary
The practical lesson from this episode is not about Fable 5 specifically. It is about a structural change in how frontier AI is governed. For the first time, the US government exercised direct control over who could access an AI model — not through terms of service but through export law. Non-US users had access removed for three weeks without warning.
For UK small businesses, this matters in two ways. First, access to frontier AI tools can now be interrupted by decisions made in Washington. Second, AI companies are being pushed towards more rigorous pre-release safety evaluation, which is ultimately good for business users — it reduces the chance of deploying a model that causes a serious compliance or reputational incident.
Anthropic, Amazon, Microsoft, and Google have agreed to develop a shared scoring framework for evaluating jailbreak techniques in frontier models. If adopted broadly, this would mean all major AI providers use a common standard for assessing whether a new vulnerability requires restricting access. Expect this to become an industry norm over the next 12 months.
What this means if you use Claude in your business
Most UK small businesses using Claude will notice no service disruption going forward — the 50% usage limit for early July is a temporary phase-in measure. The episode is more of a signal about planning than about immediate operational risk. Build your workflows knowing that access to specific model versions can be interrupted by factors outside your AI provider's control. Where possible, build adaptable workflows that can fall back to an alternative model version if a specific model becomes temporarily unavailable.
Operator actions
Today: Log into claude.ai and confirm you have access to Fable 5. If you were using Claude for document or code-related tasks during the June 12–July 1 window, review whether any workflows need updating now that Fable 5 is restored.
This month: Note the common jailbreak framework development as a trend to follow. AI safety governance is moving from company policy to shared industry standards with government input — this will affect procurement decisions and AI compliance policies for UK businesses later in 2026.
By Q4: If you are building AI workflows, include model-version flexibility as a design requirement. Avoid hard-coding specific model versions into critical business processes.