The Export Ban That Wrote Its Competitors' Marketing Copy

There’s a particular kind of own goal where you not only miss the net, you kick the ball directly to the opposition striker. Washington’s AI export controls are shaping up to be one of those.

The short version: Anthropic and others pushed for export restrictions on frontier AI models, ostensibly for safety reasons. Within weeks, labs in Tokyo and Beijing were shipping products with a pitch they could not have written better themselves. Sovereign AI. No export control risk. Your access won’t vanish overnight because someone in Washington had a bad morning. That wasn’t a compelling sales angle before the ban. The ban made it one.

One person in the discussion put it well: you can’t buy that kind of opening. The restriction manufactured the demand it was trying to contain. Classic.

There’s a legitimate counter-argument, and I want to name it rather than wave it away. Safety concerns in AI development are real. The alignment problem is real. Deployment speed without scrutiny can cause genuine harm. I believe that. The trouble is that “safety” and “protecting market position” are not mutually exclusive motivations, and when the latter is clearly in play, it corrodes the credibility of the former. Anthropic is a company with commercial interests. So is OpenAI. Regulatory capture dressed in safety language is still regulatory capture.

The copyright dimension of this is its own mess. The argument doing the rounds is that Chinese labs distilled Claude and GPT-4 without permission, essentially training cheaper models on the outputs of the expensive ones. Anthropic is reportedly raising this with Congress. Which is a genuinely interesting position for a company that trained its model by scraping the entire internet, including work by people who never consented and never got paid. Someone in the comments made exactly this point and it’s hard to dismiss. The rules seem to matter a lot more when you’re the one they’re being applied to.

I should say: I don’t know enough about the specific technical claims here to have a firm view. Some of the “Mythos-like” framing in the original reporting sounds like marketing rather than engineering. A few people in the thread flagged that some of these systems are orchestrators running on top of existing models, including Claude, rather than genuinely independent frontier models. That’s a meaningful distinction. Orchestrating multiple SOTA models simultaneously will obviously produce strong results; that’s not the same as having built a competitive base model from scratch. The headline doesn’t always reflect the reality underneath it.

What I do think is true: the geopolitical framing of AI as a race to be won by one side is probably not a useful frame. It tends to produce exactly the kind of heavy-handed policy that backfires. And the people who get locked out first are rarely the ones the policy was designed to protect. Not the big labs. Not the governments. The researchers, the small developers, the public institutions trying to work with these tools in good faith.

We spent thirty years telling the world that free markets and open competition were the answer to everything. Now that someone has shown up to compete seriously, the response is licensing regimes and review boards. I understand why. I’m not sure it’s going to work.

The honest position is probably this: some regulation of frontier AI makes sense, the current approach looks more like incumbency protection than safety policy, and the world is going to develop these models regardless of what Washington decides. Those three things are all true at the same time and they don’t resolve neatly into a clean take.

I’m not sure anyone in this space actually has a plan. That’s the part that keeps nagging at me.