Carney at G7: Fable 5 ban is "model risk" — the same systemic trap as 2008, and Canada's $2.3B answer to it

Canadian PM Mark Carney — a former Bank of England and Bank of Canada governor who navigated the 2008 financial crisis — used Day 2 of the G7 summit in Évian to name what the Fable 5 shutdown really is: "model risk," the same concentration danger that nearly collapsed global banking when Lehman fell.

"The situation we're in collectively right now with Mythos and Fable is something that can happen with over-reliance on certain models," Carney said, per The Next Web. "Nobody's done anything wrong in this situation, but we will have done something wrong if we just accept this, don't take the lesson, don't build out and diversify." He directly compared the dynamic to the systemic linkages that made 2008 so severe: concentrated dependencies that look fine — until a single node fails and everything connected to it goes down with it.

The 2008 parallel is precise, not rhetorical. In 2008 the problem was that interconnected counterparty risk meant a failure at one institution propagated everywhere. Here: a Commerce Department export directive against a single company's two model IDs effectively severed a chunk of global AI infrastructure with 90 minutes' notice, no public process, and no established fallback. Developers who had built production workflows on claude-fable-5 or claude-mythos-5 simply stopped working. The concentration risk wasn't theoretical — it was realized.

Canada's policy response. Carney used the moment to announce Canada's $2.3 billion "AI for All" national strategy, which focuses specifically on sovereign computing infrastructure — GPU capacity, training facilities, and inference resources that aren't dependent on any single US vendor or US regulatory decision. "It is never a good idea to have one option," he said, per Decrypt.

For builders: this is a supply chain problem, not a policy problem. The regulatory environment will not stabilize before your next production deployment. What Carney named — "model risk" — is real vendor dependency risk, the same category as "we only have one database" or "our payments go through a single processor." Practical mitigations now:

• Prefer versioned, non-alias model IDs and abstract them behind an env var or config so a swap is a one-line change
• Test a secondary provider on your most critical workflows quarterly — not just for performance, for operational continuity
• Track whether your vendor has any regulatory exposure that could trigger a sudden access withdrawal

Tomorrow (June 17) the G7 holds its dedicated AI working lunch — the first session where Altman, Amodei, and Hassabis sit across from G7 heads of state. Watch the post-summit communiqué for any language about "frontier model pre-clearance" or "coordinated evaluation frameworks" — that framing, if it appears, is the seed of future compliance burden for teams shipping AI-integrated products.

Sources: The Next Web: Carney compares Anthropic shutdown to 2008, OpenTools: Carney AI over-reliance G7, Decrypt: Canada urges AI diversification