Government-Gated AI: Washington Switches a Frontier Model Off — Then Decides Who Gets It Back

On 12 June the US Department of Commerce ordered Anthropic to restrict its two most capable models, Fable 5 and Mythos 5, to US persons only — and because the company cannot reliably sort users by nationality, the practical effect was a global kill switch that took both models offline for everyone within hours. Why this is a first: export controls have long covered software and source code, but this is the first time they have been used to control access to a continuously available, API-served AI model — not a chip, not a file, but a capability reached over the cloud. The trigger is contested: the order followed a narrow jailbreak of Fable's cyber guardrails reported by Amazon researchers; Anthropic says the same capability is widely available in other deployed models — including OpenAI's GPT-5.5 — and at least one researcher who saw the work disputes the "jailbreak" framing. Then the gate reopened, selectively: on 26 June, Commerce re-authorised Mythos 5 for roughly 100 trusted US organisations and agencies, while OpenAI previewed its new GPT-5.6 Sol only to partners the government had individually vetted — two labs, one Commerce Department, one emerging template of "managed-release" AI, with Fable 5 still dark. Why this matters for ship managers: shipping is one of the most multinational industries on earth — mixed-nationality crews, offices and operators — so a "no foreign nationals" rule on the best models is structurally awkward for this sector in particular, and a growing share of maritime tooling (enterprise assistants, the CMA CGM–Mistral alliance, dozens of API-built vendors) now sits on top of frontier models that can be switched off by directive overnight. The fallout was immediate — a surge toward cheaper Chinese open-source models, a record DeepSeek raise, and an AI-sovereignty scramble in Europe. The practical takeaway is continuity: know which models your AI suppliers depend on, and whether they have a fallback if access changes. [CSIS · Lawfare]

This story is part of the Maritime AI Digest of 28 June 2026.