When a directive overseas can switch off your AI overnight

Over the weekend, access to one of the most capable AI systems in the world changed hands on the basis of nationality, not contract. As SmartCompany reports, a US national-security directive suspended foreign-national access to Anthropic’s frontier models — the general-purpose Fable 5 and the cyber-focused Mythos 5 — cutting off Australian businesses, including participants in the Project Glasswing programme.

Anthropic disputes the technical basis for the directive. For an Australian business that woke up on Monday unable to reach a model it had built around, the dispute is academic. The practical reality is the part worth sitting with: access to a leading model can now turn on geopolitics rather than commercial terms. You can be a paying customer in good standing and still be switched off by a decision made in another country, for reasons that have nothing to do with you.

This is not really an AI story

It is tempting to read this as a story about Anthropic, or about one government’s policy. It is more useful to read it as the clearest platform-risk wake-up call the AI era has produced so far.

Platform risk is an old idea. Businesses that built their entire customer base on a single social platform learned it when the algorithm changed. Businesses that ran on one payment provider learned it when an account got frozen. The pattern is always the same: you build something valuable on top of infrastructure you do not control, it works beautifully for a while, and then the terms change without your input.

What is new is the speed and the leverage. A frontier AI model can sit at the centre of how a business drafts, classifies, summarises, and decides. When that much of an operation depends on one provider, a single directive can degrade it overnight. The question this weekend put to every business using AI is simple: if your most important model disappeared on Monday, what would still work?

Model-agnostic is the boring answer that holds up

The way you survive this is unglamorous: do not hard-wire your operation to one model.

At func.digital we build automation on composable, model-agnostic systems. In plain terms, the workflow — the enquiry triage, the quote follow-up, the reporting — is designed so the AI model is a component you can swap, not a foundation you are welded to. The logic of “classify this enquiry, draft a reply, update the CRM” does not care which model does the classifying. If one provider becomes unavailable, too expensive, or geopolitically off-limits, you change the component and the workflow keeps running.

That is the difference between an automation that survives a weekend like this one and an automation that simply stops. The first treats the model as a swappable supplier. The second treats it as load-bearing, and load-bearing things are exactly what you do not want a foreign directive able to remove.

The conversation to have now is continuity, not cleverness

For most of the last two years, the AI conversation has been about which model is smartest this month. This weekend reframed it. The more important questions for an Australian business are about continuity and sovereignty:

• Single points of failure — where in your operation does one AI provider sit on the critical path, with no fallback if it goes dark?
• Switchability — if you had to change models tomorrow, is that a config change, or a rebuild?
• Data residency and control — where does your data go, under whose jurisdiction, and what happens to your access if that jurisdiction’s politics shift?
• Graceful degradation — when the model is unavailable, does the workflow fail safely and route to a human, or does it just break?

None of these are questions about which model writes the nicest email. They are questions about whether the business keeps running when something outside your control changes.

What to check this week

You do not need to rip anything out to get a read on your exposure:

1. List the AI that is actually in production. Not the experiments — the things customers or staff now depend on. For each, name the provider.
2. Mark the single points of failure. Anywhere one provider has no alternative is a place a directive, an outage, or a price change can hurt you.
3. Ask “swap or rebuild?” for your most important one. If the honest answer is “rebuild,” that is the workflow to make composable first.
4. Decide your fallback. For each critical automation, what is the plan when the model is unavailable — a second provider, a simpler model, or a human in the loop?

The func.digital take

We are not arguing you should avoid frontier models. They are genuinely useful and you should use the best one available to you. We are arguing you should not build your business so that losing one of them is a crisis.

Build the workflow to outlast the provider. Keep the model swappable, keep a human able to step in, and keep your data somewhere you understand. Then a weekend like this one is a headline you read with interest, not an outage you live through. If you are not sure how exposed your operation is, that is exactly what a digital systems audit is for — we map where the single points of failure are and what it would take to make them swappable. Get in touch before the next directive, not after.