The US Just Proved It Can Switch Off Your AI Overnight

On June 12, the US Commerce Department issued an export control directive to Anthropic — the company behind the Claude AI assistant. Within hours, Anthropic disabled its two most advanced models, Fable 5 and Mythos 5, for every user on the planet. No warning. No transition period. Businesses that had integrated those models discovered the tools had stopped working at 5:21 PM US Eastern Time.

The stated trigger was a narrow jailbreak vulnerability that Anthropic described as already present in competing models. The real story is bigger than one vulnerability: the US government just demonstrated it can switch off commercial AI tools globally, overnight, for reasons of its own choosing.

This was not a chip embargo or a hardware restriction. It was the first time Washington applied export controls to a finished AI product that businesses use daily. The directive required Anthropic to block all foreign nationals from accessing the models — not just in sanctioned countries, but everywhere, including foreign-born employees working inside the United States.

Because cloud AI providers know where their users are but not what passport they hold, Anthropic could not selectively comply. It disabled both models for all customers globally. Anthropic’s other Claude models remained available, but the two most capable did not. As of June 20, access has still not been restored. Anthropic says it expects to introduce identity verification in early July, after which access should resume for allied nations including Australia.

Roy Morgan research published this month shows 13.6 million Australians — 58 per cent of the population aged 14 and over — now use AI tools in a typical month. The three most popular platforms are ChatGPT (used by 10.5 million Australians), Google Gemini (5 million), and Microsoft Copilot (4 million). All three are built and controlled by US companies, subject to US law and US government directives.

Last week’s directive hit Anthropic, which has a smaller Australian user base of around 777,000. But the precedent applies equally to OpenAI, Google, and Microsoft. If Washington issues the same directive to any of those companies tomorrow, the tools that millions of Australian workers depend on could go dark with identical speed. The only major AI tool in Australia’s top five not controlled by a US company is Canva Magic Studio — built in Sydney.

Ten weeks before the export control landed, the Australian Government signed a memorandum of understanding with Anthropic to collaborate on AI opportunities. Anthropic’s CEO Dario Amodei met with Prime Minister Albanese to formalise the agreement — the first signed under Australia’s National AI Plan. We’ve written before about Canberra’s AI regulation approach. Last week exposed a gap that no domestic framework can close: the tools themselves are controlled offshore.

Consider a law firm using Claude to review contracts, or a trades business running AI-powered scheduling built on a US platform. When the tool goes dark, the work does not stop — but it slows dramatically. The firm burns senior associate hours on tasks the AI was handling. The trades business reverts to manual dispatch and loses the optimisation that was adding two or three extra jobs per day.

If your team uses AI for quoting, document drafting, client communication, or scheduling, you have an operational dependency on a foreign-controlled service. That dependency worked fine until last week. Now there is a precedent showing it can be revoked without notice. This is not an argument against using AI. The productivity gains are real. But treating an AI tool as if it were as reliable as electricity — always on, always available — is no longer a defensible assumption.

First, map your AI dependencies. Which processes rely on a specific AI tool? If that tool disappeared tomorrow, what breaks? Most businesses have never asked this question because the tools have always been there. That is exactly the point.

Second, avoid single-provider lock-in where it matters. The major AI providers’ APIs are increasingly interchangeable for common business tasks. If your critical workflows can run on more than one platform, a disruption becomes an inconvenience rather than a crisis.

Third, keep a manual fallback for your highest-value processes. Not because AI will fail often — it probably won’t. But because the decision to switch it off may not be yours to make.

Key takeaways

Want to understand your AI dependencies before they become urgent?

The right time to map your AI exposure and build resilience is before the next directive — not after. A short conversation can identify where your business is vulnerable and what to do about it. Book a call.