Video: "Claude's Fable 5 Just Got BANNED!" by Julian Goldie on YouTube.
What the export control order actually says
The US Department of Commerce issued an emergency export control directive on 12 June 2026, three days after Anthropic launched Claude Fable 5 and Mythos 5. The directive places both models on the restricted list under export administration regulations, barring access for all foreign nationals — which in practice means anyone outside the United States. Anthropic complied immediately, pulling access for non-US accounts the same day.
The order is model-specific. Older Claude models — Opus 4.8, Sonnet 4.6, Haiku 4.5, and the earlier Fable family — are not named and remain fully available. This is an emergency measure rather than a permanent rule change, but there is no stated timeline for review or reversal.
Why the government acted
Anthropic's own published safety evaluations described Mythos 5 as "far ahead of any AI in cyber capabilities." The evaluation documented the model's ability to find zero-day vulnerabilities autonomously, without human prompting — the kind of capability that triggers national security review under existing dual-use technology rules. Claude Fable 5 was built from the same underlying base, which is why it was included in the same directive despite a different capability profile.
This is not the first time powerful AI has intersected with export controls, but it is the first time the US has moved this quickly — within days of a public launch rather than months after a capabilities assessment. The speed of the intervention itself tells you something about how seriously the capabilities were taken internally.
Which Claude models UK businesses can still use
Opus 4.8, Sonnet 4.6, and Haiku 4.5 are all unaffected. These are still Anthropic's best publicly available models for the vast majority of business tasks — code, content, analysis, agentic workflows, customer-facing automation. Claude Code runs on Sonnet 4.6 and Opus 4.8 by default, and both remain fully accessible to UK users.
Hermes Agent setups that use Claude through the API continue to work without changes. If you built a workflow around Fable 5 specifically — it was only publicly available for a matter of days — you'll need to switch the model parameter to Opus 4.8 or Sonnet 4.6. In most cases the difference in output quality will be modest for everyday tasks.
What this means for AI agents and automation
For most businesses running AI agents today, practically nothing changes. The workhorse models are unaffected, and the day-to-day output of a well-configured Claude Code or Hermes Agent setup does not depend on Fable 5. The more significant concern is longer-term: what this episode shows is that even Frontier-model releases can be pulled within days of launch, without warning, for non-commercial reasons outside any business's control.
That means designing production workflows around specific model versions carries more risk than it appeared to before 12 June. The safer approach is to build agent configurations that specify a model tier — frontier, balanced, fast — rather than a specific model name, so your workflow continues working when the underlying model changes beneath it.
Where this connects to NordSys
We install and maintain Claude Code and Hermes Agent for UK businesses, and part of that work is making sure your AI setup doesn't break when Anthropic releases a new model or, as this week showed, pulls one. The configurations we build are designed to stay stable across model changes — with the right fallbacks, the right model tier settings, and the right testing before any update goes live.
If you're running a Claude-based workflow and want to make sure it's set up to handle model availability changes without disruption, that's exactly the kind of thing a care plan covers.
See our AI Agents service →