On 12 June, every Australian quietly lost access to Anthropic’s Claude Fable 5 — no warning, no negotiation; the model simply went dark after a directive from Washington that cited national security.
The shutdown: a foreign off‑switch made real
The interruption was not a technical failure or a cloud outage. According to the account in the source piece, access to Anthropic’s Claude Fable 5 was suspended for Australians the same evening a United States directive limited foreign access. The article frames that single decision as a demonstration of a structural risk: when critical services are built on foreign models, sovereignty over use can be removed by an external actor pulling an off‑switch.
How France chose a different path: Mistral and national controls
The source contrasts Australia’s exposure with France’s approach. Three French researchers from Google DeepMind and Meta founded Mistral in 2023; within two years it became “Europe’s most valuable AI company and the continent’s answer to OpenAI,” the article reports. Mistral released capable models with open weights that could be downloaded and run on domestic servers. In December 2025 the French Ministry of the Armed Forces signed a framework agreement for Mistral to supply AI services across the defence establishment — including the armed services, the atomic energy commission and the navy’s hydrographic office — subject to non‑negotiable conditions that everything run on French‑controlled infrastructure and that data never leave national jurisdiction.
Three concrete moves the article urges for Australia
The piece lays out a three‑part strategy it says would produce sovereign capacity. First: treat computing power as critical national infrastructure and define the essential systems — the source lists “core government and the payments system to grid control and defence” — that must remain active through external shocks. Second: when global tech giants build data centres here, secure a guaranteed, non‑revocable share of computing power set aside for Australian use; the article notes those tech giants are “about to spend close to a trillion dollars a year” on such capacity and argues Australia should compete for the investment while naming a price. Third: take a cornerstone equity stake in an Australian AI developer sufficient to keep model weights and decisions onshore, make the federal government the anchor customer, and contractually require agencies that handle the nation’s most sensitive data to run on a model hosted on Australian servers with guarantees that “no foreign directive can reach the off‑switch.”
How policymakers, technologists and procurement leaders will respond
- Policymakers and regulators: The article calls on the federal government to start by designating computing capacity as critical infrastructure and to use procurement and equity stakes to keep control of core models and hosting under Australian law.
- Technologists and security teams: For teams responsible for sensitive systems — the source names payments, grid control and defence — the recommendation is to require models and hosting that run on Australian‑controlled servers and to insist on contractual guarantees preventing foreign directives from disabling services.
- Procurement leaders and agency heads: The article urges these actors to follow Paris’s example by directing institutions to buy domestically hosted services; it notes France backed its demand with capital and a national AI push pledging to grow to €109 billion (A$180 billion).
Where the National AI Plan fits — and where it does not
The source critiques Australia’s National AI Plan, released in December 2025, as full of commitments about the AI “we use” — a Safety Institute, pledges to share benefits, and promises to carry workers through disruption — but lacking measures to ensure Australia will be allowed to use that AI in the first place. The article argues that while safety and workforce programs matter, they do not address the vulnerability exposed by the Claude Fable 5 shutdown: dependency without a reserve. It compares computing power to strategic fuel reserves and asserts that Australia’s reserve of computing power “sits at a proud zero.”
Conclusion — a policy choice already being made
The central observation in the source is blunt: data centres are being built regardless, and the only question is whose they are. Absent deliberate policy — treating compute as critical infrastructure, reserving capacity in commercial builds, and anchoring a domestic model and hosting — the article warns that that question will be answered “for us, cheerfully, by people who haven’t grasped that they’re answering it.” The account offers France’s Mistral as a working example of how a country can secure operational control without trying to win the frontier race for the single most advanced model. For Australia, the tradeoffs described are immediate and concrete: convenience and faster access from foreign providers, or a deliberate investment in sovereign capacity so a single foreign directive cannot render core systems unusable.
