Every major technology eventually becomes a strategic asset. It's happened with electricity, semiconductors, and now AI. A single government order last week pulled one of the most capable AI models offline worldwide, demonstrating that access can be revoked overnight. The capability you build and own cannot.
On June 12, the United States government, citing national security, ordered Anthropic to suspend access to its two newest and most capable models, Fable 5 and Mythos 5, for all foreign nationals. According to Anthropic, the concern was a method of bypassing, or jailbreaking, the safeguards that stop Fable 5 from being misused for tasks such as cyberattacks. In a public statement, Anthropic disputed both the severity of the finding and the standard it would set, since the same capability is already available in other models. Recalling one on this basis would halt new releases altogether. The United States has previously used export controls to keep advanced semiconductors out of certain countries, but according to TIME, this was the first time it applied them to an AI model itself. A model is now being treated as a strategic asset in its own right, rather than only the chips used to train it.
"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."
— Prime Minister Mark Carney
This is an early example of a broader risk, what happens when essential work depends on a capability that someone else can switch off. Prime Minister Mark Carney named that risk directly, calling it "model risk" and comparing it to the 2008 financial crisis. He argues that the discipline regulators then imposed on banks, redundancy and diversification, now applies to anyone building on AI.
Access is not capability
Carney's call to diversify points to the distinction that access and capability are not the same thing. Access is rented and can be withdrawn by external forces, while capability (the compute and models you actually own and run) cannot be taken away. Canada is making that choice at a national scale, betting on capability over access with its "AI for All" strategy, which commits $2.3 billion to building domestic AI capacity and makes sovereign compute a national priority.
Building capability at that scale will be out of reach for many companies, but they might not need to. Some organizations will have good reasons to build their own models or infrastructure. However, for most, resilience comes not from owning the AI itself but from controlling how it is built into the business, the way your systems are designed, the providers you rely on, and the fallbacks you can switch to when one fails.
What to actually do about it
The practical response is to treat model choice as an architecture decision that depends on the stakes of the work, rather than a brand commitment. Design the systems that matter to be model-agnostic, with a routing layer that can switch providers without rebuilding the workflow underneath, and put a tested fallback in place before you need it.
The market is already moving to meet this need, a day after the Fable shutdown OpenRouter released its Fusion API, which sends a prompt to several models at once and synthesizes the results, so that no single model failing can break the result. Tools like this turn the routing layer into something you can buy rather than build, which lowers the barrier for smaller teams. The trade-off is that you are now depending on the aggregator instead of the lab. It reduces single-model risk, but doesn't remove dependence altogether. This belongs in your toolkit alongside alternate providers or self-hosted fallbacks rather than completely solving the problem.
For most BC businesses the immediate impact of the Fable shutdown was small, however, the risk it exposed is real and growing. The better these models get, the more work will run through them, and the more a sudden loss of access will cost. Renting access has looked more efficient than building capability for years, and for most workloads it still is, but the Fable shutdown is a reminder that renting comes with terms you do not set. The takeaway is not to slow down on AI. It is to build it in a way you can rely on, so that no single model, vendor, or government decision can take your business offline. For BC leaders, it's worth evaluating whether your organization can keep running if any one of these pieces were to fail.
Adoption by itself cannot be expected to provide much of a durable advantage. But capability you build and own, that is what will stick around.