What the 'AI Strait of Hormuz' Means for Your Business

Last Saturday I wrote about Claude Fable 5's pricing. By Sunday the same model was at the centre of a sovereignty argument.

Over the weekend of 13 and 14 June, a US government directive suspended foreign-national access to Anthropic's two newest models, Fable 5 and Mythos 5, on national-security grounds. Developer Simon Willison documented the directive as it landed, and by Monday the Australian Financial Review's Paul Smith had the local reaction under the headline "Fears of 'AI Strait of Hormuz' as Trump's Anthropic ban shocks business".

The strait metaphor came from Employment Hero's Ben Thompson, quoted in that AFR piece: like oil tankers through a chokepoint someone else controls, Australian businesses consume frontier AI, we do not build it, and the flow can be shut off from Washington. Former national cyber security adviser Alastair MacGibbon put it more sharply in the same article: a foreign government can switch off a tool our hospitals and utilities are being urged to rely on, at a few hours' notice, with no consultation.

For a decision that big, the notice given to paying customers outside the US was effectively zero.

The uncomfortable question for operators

Strip away the geopolitics and the story leaves one question on every operator's desk: if your primary AI provider went dark tomorrow morning, what would stop working in your business?

For plenty of businesses in 2026 the honest answer is "more than we thought." Quoting workflows, customer-service triage, document drafting, code assistance. Individually small dependencies, accumulated quietly over eighteen months, mostly on a single vendor because standardising on one provider was simpler.

I run my own firm substantially on Anthropic's models, so this is not finger-wagging from a distance. The directive was a live test of my own continuity thinking, and I did not enjoy every answer I found.

What resilience looks like at mid-market scale

The AFR piece carries the standard enterprise prescription, again from Thompson: stay model-agnostic across multiple providers, and for larger players, self-host open-weight models. He also flags, correctly, that most SMEs lack the talent and capital to self-host. So what does the practical version look like for a business of 30 to 300 people?

Know your dependency map. One page. Every workflow that touches an AI model, which provider it runs on, and what the manual fallback is. Most businesses have never written this down, and it takes an afternoon. You cannot manage a concentration you have not measured.

Own your prompts, data and workflow logic. The switching cost between AI providers is mostly in the scaffolding: the prompts, templates, reference documents and process definitions built up around the model. Keep those in your own systems, in plain formats, rather than locked inside one vendor's platform. If the scaffolding is portable, the model underneath becomes swappable.

Test a second provider before you need it. Not a full migration. Pick your single most AI-dependent workflow and run it through an alternative model once a quarter. You are buying two things: proof the fallback works, and a realistic sense of the quality gap. A fallback you have never tested is a hope, and hope is a poor continuity plan.

Match the effort to the stakes. If AI drafts your internal meeting notes, an outage is an annoyance and deserves no engineering. If it sits inside how you quote, serve customers or ship product, it deserves the same continuity treatment as your accounting platform or your internet connection.

Skip the panic self-host. Running open-weight models on your own infrastructure is a serious engineering commitment. For most mid-market businesses the resilience-per-dollar is far better in portability and a tested second provider than in building a private AI stack because one weekend's headlines were frightening.

The wider Australian picture

The same AFR article notes former industry minister Ed Husic's criticism that Australia's planned National AI Capability Plan was diluted into a National AI Plan without a clear roadmap, while the EU works to reduce its dependence on US providers. Whatever your politics, the practical reading for an operator is that no sovereign backstop is coming on any timeline that helps you. Resilience, at our scale, is self-service.

Here is the framing I would take to your next leadership meeting. We buy electricity from a grid we do not control, and we manage that risk with contracts and contingencies rather than by refusing to use electricity. AI has just announced itself as the same class of dependency. The businesses that respond by pulling back from AI will pay a competitiveness price for an availability risk they could have managed for a few thousand dollars of effort. The businesses that respond with a dependency map, portable scaffolding and a tested plan B get to keep compounding the benefits, with an exit route if the strait ever closes properly.

One afternoon for the map. Start there.

Where does your business stand? The free AI Scorecard takes three minutes and shows you. If you want a straight steer from a person, book an AI Opportunity Call.