The foreign national label: reaching a frontier model from Turkey

Last week a model disappeared from my subscription plan without me doing anything. The reason had nothing to do with my bill, my usage, or even Anthropic. The reason was the color of my passport.

Within two days I had written about Fable 5 going public and then about that same model being pulled by a government directive. I reported the event but didn't sit with it. A few days later I realized the real story isn't one model being withdrawn — it's the ground that access stands on.

Access is a permission, not a right

We solo developers think: "I have an API key, I pay my bill, therefore I have access to the model." The Fable 5 episode quietly broke that assumption. The directive's scope wasn't "abusive users"; the scope was the category foreign national. Whether you live inside the US or outside, whether you pay on time — none of it matters. The line wasn't technical; it was legal and geographic.

For someone working from Turkey, this isn't an abstract debate. The "foreign national" that directive defines is precisely me. I was in scope not for misusing the model, but simply for standing on the wrong side of a border.

Why it suddenly became visible

For a few years there had been an asymmetry in frontier-model access, but it was invisible. ChatGPT didn't launch in some countries; some APIs were closed to certain regions. We read those as "they haven't entered the market yet." With Fable 5, for the first time the rationale was written plainly as national security and export control.

The difference matters: market decisions can change — a company could open Turkey tomorrow. But an export control isn't at a company's mercy; it's in a government's pen. Anthropic didn't want to pull the model; it had to. That shows where the power actually sits.

The practical takeaway for a solo builder

The lesson I took from this isn't pessimism — it's architecture. The cost of locking into a single provider, a single model, is no longer just "a price hike"; the scenario "your access is fully cut off one morning" is now on the table too.

In practice:

• I design critical flows to be model-agnostic; I try to make the provider a single config line.

• I keep at least one backup provider (from a different jurisdiction) actually tested — not "I'll look into it if needed," but "it works today."

• I now see open-weight models not as a "second-class option" but as access insurance. I had already written that open models are eating the market; this episode also vindicated that thesis on a political axis.

Limits / concerns

It's easy to overstate this. The Fable 5 case rested on a narrow, specific security rationale; it doesn't mean every model will be pulled tomorrow. Opus 4.8, Sonnet, GPT-5.5 — all kept running. So awareness, not panic, is the right response.

But once that door is open, you have to assume every new model that fits the "frontier" definition is exposed to a similar risk. The most capable model is usually the one most likely to be restricted.

A note from me

Honestly, this hit me harder than a technical headline. For years I've worked on the assumption that "the internet is borderless, just learn and build." Fable 5 reminded me that assumption has a footnote: the borderless thing is knowledge; the tools have borders.

It doesn't stop me — if anything it makes me more disciplined about building to stay independent. If there's a cost to building from Turkey, I'd rather absorb that cost in my architecture than stop building.