AI Sovereignty Alarm: Carney Compares Fable 5 Shutdown to the 2008 Financial Crisis

On the evening of June 12, a single US Commerce Department export control order pulled Anthropic’s two most powerful models, Fable 5 and Mythos 5, off every screen worldwide at once. Three days later, on the sidelines of the G7 summit in Évian, France, Canadian Prime Minister Mark Carney put it in blunt terms: this is the same systemic logic that drove the 2008 financial crisis.

Why Carney can make that claim stick

Coming from most politicians, that line would be theater. Coming from Carney, it carries weight. He ran the Bank of Canada, then spent six years running the Bank of England, where his entire job was watching for concentration and fragility in the financial system. He raised the comparison with reporters during a visit to Ireland, then brought it straight to the G7 table, taking it up directly with French President Emmanuel Macron.

The phrase he reached for was “model risk.” His exact words: “We have similar things in terms of model risk.” He followed with a sharper point: nobody did anything wrong in this particular episode, but failing to act on the lesson would itself be the mistake.

In finance, “model risk” describes what happens when institutions lean too hard on a single model or a single set of assumptions, then fail together the moment reality moves outside the tested range. Carney is pointing the same structure at AI: when global enterprises route critical workflows through a handful of models from a handful of US vendors, a single directive is enough to sever the whole chain at once. Within a day of the Commerce Department’s order taking effect, Fable 5 and Mythos 5 users everywhere lost access — Carney’s timing on the analogy was no accident.

Canada already placed its bet

This wasn’t an off-the-cuff political moment. On June 4, Carney’s government unveiled “AI for All,” a national strategy committing more than $2.3 billion in spending over five years, with one pillar explicitly dedicated to building sovereign Canadian compute infrastructure. Of that, up to $1 billion goes toward a public supercomputer, giving Canadian researchers and businesses a path that doesn’t run entirely through US cloud providers. The strategy also sets two job targets for 2031: up to 90,000 AI-related positions and placements for young Canadians, plus 250,000 broader positions created through AI adoption across the economy.

The export order simply handed Carney’s three-week-old policy decision a ready-made case study.

From Canada to the UK and India, sovereign compute is surging everywhere

This isn’t a Canada-only push. On June 8, UK startup Cosine lined up more than a dozen major firms, including BT, HSBC, Lloyds, NatWest, and BAE Systems, to back “Lumen Sovereign,” a frontier model trained entirely on Britain’s own Isambard-AI supercomputer. The goal is explicit: regulated sectors like finance and defense shouldn’t have to hand both their data and their compute to American companies.

India’s reaction has been even more direct. The moment Fable 5 went dark, local outlets framed it as handing India’s sovereign AI movement its strongest argument yet. The country already has more than 1,700 AI-native companies that have collectively raised roughly $5.5 billion, and a long-floated proposal for an annual $5 billion sovereign AI fund is back on the table because of this episode.

Three countries, three different playbooks, the same underlying logic: nobody wants their industry’s computing lifeline to depend on another country’s single executive order.

Not everyone is buying the comparison

Carney’s framing has plenty of skeptics. AI researcher Gary Marcus put it bluntly, calling the export controls a case of the US “self-sabotaging” its own AI competitiveness. Investor Martin Varsavsky offered a more practical warning: once this kind of control sets a precedent, it rewrites the rules for the entire industry, not just for Anthropic.

Anthropic itself hasn’t budged. The company maintains the vulnerabilities that triggered the order were minor, and that rival models carry similar weaknesses — meaning it was blocked by a safety standard the industry hasn’t actually agreed on.

What to watch next

A $2.3 billion commitment is easy to announce. Turning chips, power, and talent into an actual working supercomputer within a useful timeframe is a different order of difficulty entirely. As the G7 summit runs through its remaining two days, expect more countries to put similar sovereign-compute plans on the table. The number worth tracking isn’t how many more billions get announced — it’s how much of that money actually turns into usable compute capacity.

If this was useful, subscribe to the newsletter for weekly AI PM insights and GenAI case studies.

Sources: The Next Web, Prime Minister of Canada, Tech.eu

---

Related Reading:

• US Government Forces Fable 5 Offline: How One Jailbreak Demo Took Down Anthropic’s Flagship Model
• G7 Summit 2026: Altman, Amodei, and Hassabis Meet World Leaders Together for the First Time