On 13 June, Anthropic switched off its two strongest AI models, Fable 5 and Mythos 5, for every user outside the United States. The order came from the US government, not the company. An export-control directive from the Trump administration told Anthropic to block access by any foreign national on national security grounds. That includes South African businesses, South African developers, and even Anthropic staff who are not US citizens.

If you were using either model in a tool or workflow, it stopped working that day. No phase-out, no migration window.

Why It Happened

The trigger was a jailbreak. Anthropic said the government had found a way past Fable 5's safety controls, and that the model was unusually good at one specific thing: finding security vulnerabilities in software. A model that can scan code and surface exploitable weaknesses at scale is useful to defenders and just as useful to attackers. Washington decided the second risk outweighed the first and acted before the method spread.

Anthropic complied the same day. It said it had received only verbal notice of a narrow, non-universal jailbreak, which tells you how fast this moved.

What Changed Since

On 27 June the government cleared Mythos 5 for limited use again, but only for a small group of cyber defenders and infrastructure providers inside the US. Fable 5 is still restricted. For anyone in South Africa, both are off the table for now.

What This Means for South African Businesses

Two practical points. First, if your AI tools quietly depend on one specific top-tier model, you are exposed to decisions made in another country. Build your workflows so you can swap the model underneath without rebuilding everything around it. Second, the frontier models are now a geopolitical asset, and South Africa sits outside the circle that gets first call.

The reassuring part: the models most businesses actually use day to day, the standard Claude and GPT tiers, are still available here and still very capable. You do not need a vulnerability-hunting research model to run a customer chatbot or draft proposals. Match the tool to the job, keep a fallback, and do not build your business around a single model you cannot control.