Ask who benefits before you ask who is right

Two lobbying pushes hit the same week, aimed at each other. Axios reported that OpenAI and Anthropic are pushing the US government toward restricting open weight AI, framed as a national security concern. Politico reported the opposite pressure from the other direction, roughly two hundred startups under something called the Little Tech Association asking the same government not to cut off Chinese open weight models, on the grounds that plenty of small companies are building on them right now.

Notice that both sides are making a policy argument, and both sides also happen to want the policy that helps their business survive. That is not an accusation, it is just how incentives work, and pretending otherwise is naive in either direction. A closed frontier lab genuinely believes open weights carry misuse risk, and also genuinely benefits if regulation locks in its advantage. A startup running on a cheap open model genuinely believes access matters, and also genuinely needs that access to keep its costs down.

The vault’s notes on this call it access volatility, the fact that which models you are even allowed to use keeps shifting under builders for reasons that have nothing to do with the model’s quality. That instability is the real cost here, more than whichever policy wins. A company that architects itself around one model or one country’s export rules is betting on a fence that a lobbying campaign might move next quarter.

So when you read either argument, read it twice. Once for the stated reasoning, once for the balance sheet standing behind it. Usually you will find both are true at the same time, and that is exactly what makes the debate so hard to referee honestly.