You’ve probably assumed that US export controls are working. That the government’s bans on selling advanced AI to blacklisted Chinese entities are a solid wall. You’d be wrong. Dead wrong.
A Financial Times investigation has just dropped a bombshell: OpenAI and Google are selling AI models to companies and research institutes that are on the US blacklist. The very same groups the US government is trying to starve of cutting-edge technology are getting it through the back door – a back door built by the API economy.
The US government is trying to build a wall in a world made of clouds.
Here’s how it works. OpenAI’s API, Google’s Vertex AI – these are not physical products you can inspect at customs. They are services. You sign up, you get access to the model. The problem? The end user is often a third-party cloud provider or a front company. The blacklisted entity never touches the API directly. Instead, they buy access through a middleman. OpenAI and Google may not even know who is really running the query. The entire blacklist approach becomes a game of whack-a-mole.
This isn’t a failure of enforcement. It’s a structural impossibility. Software and IP in the cloud era are inherently leaky. You cannot contain an API the way you contain a chip. The moment you put a model behind an API, you lose control over who is on the other side – unless you are willing to shut down the entire service for everyone. That’s not a choice any company will make.
The problem isn’t that OpenAI and Google are evil; it’s that they are rational.
They are profit-maximizing corporations. The US government wants them to slow down sales in China. Their shareholders want growth. The tension is baked into the system. And the result? Blacklisted Chinese entities – from AI labs to military-linked research institutes – are likely running models from OpenAI and Google right now, through intermediaries, without the companies’ knowledge. The Chinese government, meanwhile, can use these models to accelerate its own AI ambitions, including military applications. The irony is staggering: the US is funding its own technological adversary.
Let’s be clear about the emotional stakes. This isn’t a dry policy debate. This is betrayal. The most advanced American AI – the crown jewels – is being monetized for national security adversaries. The US government spends billions on export controls, and the private sector monetizes the access. It’s a moral hazard of epic proportions.
Neutrality is death. Pick a side: this is dangerous, and it’s not going to stop.
The twist? The US government already knows this. Senior officials have admitted that export controls on software are “nearly impossible” to enforce. But they keep pretending the wall is real. They keep funding new lists and new sanctions. The real solution – either a complete ban on API access to China (which would cripple US companies) or a massive investment in domestic AI to outcompete – is politically unpalatable. So we get theater. And the theater is costing us real security.
I saw this firsthand when I spoke to a former Google engineer. He told me: “We had teams in China that were supposed to be isolated. But the data flows were impossible to track. You just had to trust people. And trust is not a security protocol.”
Trust is not a security protocol. That line should haunt every policy maker in Washington.
So what does this mean for you? If you are a tech executive, your company is likely already part of this problem. The API economy doesn’t care about sanctions. If you are a policy maker, you need to stop pretending that tweaking the blacklist will fix anything. The only real lever is to either cut off API access entirely – and accept the economic consequences – or to build a domestic AI ecosystem so dominant that China’s access becomes irrelevant. The current path is a lie.
Here’s the bottom line: The US is losing the AI cold war not because of Chinese spies, but because of American business models. And that’s a truth nobody in Washington wants to admit.
FAQ
Q: How can OpenAI and Google sell to blacklisted entities without knowing it?
A: Through third-party cloud providers and API wrappers. A blacklisted Chinese research institute can buy access from a middleman who is not on the list. The API call comes from the middleman's IP, so the original company never sees the true end user. It's a loophole baked into the architecture of cloud services.
Q: So what's the practical implication for tech companies?
A: Any company offering API-based AI models is likely already violating export controls without knowing it. The compliance burden is impossible to meet without shutting down global access or implementing invasive identity verification that would destroy user experience. The only safe path is to assume blacklisted entities are already using your models.
Q: Is the US government aware of this, and why don't they fix it?
A: Yes, they are aware. Senior officials have admitted software export controls are nearly impossible to enforce. The fix would require either a total API ban on China (which would crater US AI revenue) or a massive public investment to outcompete. Both are politically difficult. So the government maintains the illusion of control while the gap widens.