In June, Anthropic launched its newest frontier model, released in two versions: Claude Mythos 5 and Claude Fable 5. Three days later, a US government export-control order meant access to both was suspended. Overnight. Worldwide.
Not because it failed. Because it became useful enough to be dangerous, and a national security order judged that restricting it served the national interest better than allowing it.
For the businesses that had built critical work around it, this wasn’t abstract. Reports stopped generating. Automated processes stalled. Teams who’d come to rely on it were suddenly locked out, with no warning and no fallback.[1] Access to Fable 5 has since returned, but Mythos 5 remains limited to a small group of US organisations, and nobody outside that group yet knows when, or whether, it will come back.[2][3]
(Update: in September, the EU’s cybersecurity agency, ENISA, confirmed it had been granted access to Mythos 5.[4])
We’re watching governments everywhere quietly trade rules-based consistency for self-interest, and AI is not going to be exempt from that.
Why financial services should care
For financial services, regulation (DORA in the EU, and the PRA and FCA operational resilience and outsourcing rules in the UK) already requires you to understand and manage the risk of relying on outside suppliers, the same principle that applies if your payroll system or your CRM went down.[5][6][7] Salesforce and Oracle are dependencies too, but they’re stable and predictable. AI is different, because AI’s own unpredictability gives governments a legitimate excuse to restrict it, with little warning and no negotiation.
That’s precisely why contingency can’t be an afterthought.
The organisations pulling ahead aren’t assuming their AI tools will always be there. They’re asking now: if this stopped working tomorrow, do we have another option ready? A smaller, self-hosted model? A European alternative? Have we actually tested what switching looks like, or are we just hoping it never happens?
If the AI tool your business depends on disappeared overnight, would you have a plan? Or would you be starting from scratch?
If you would like to test how dependent your business is on its AI suppliers, and what a sensible fallback looks like, that is a conversation I would very much welcome. You can find out more about how we work, or contact me directly.
This article was first shared as a post on LinkedIn.
Sources
- Anthropic, Statement on the US government directive to suspend access to Fable 5 and Mythos 5, 12 June 2026.
- Anthropic, Redeploying Claude Fable 5, 30 June 2026 (models released 9 June; export controls applied 12 June; Fable 5 restored globally from 1 July; Mythos 5 restored only for a set of US organisations).
- The Record, US lifts export controls on Anthropic’s frontier cybersecurity AI models, 1 July 2026.
- Silicon Republic, EU finally gets its hands on Anthropic’s Mythos, 10 September 2026.
- Regulation (EU) 2022/2554 on digital operational resilience for the financial sector (DORA), EUR-Lex.
- Prudential Regulation Authority, SS2/21 Outsourcing and third party risk management.
- Financial Conduct Authority, PS21/3 Building operational resilience.