Anthropic Got Banned, OpenAI Got the Contract, Indonesia Got the Trade Deal

Anthropic, the AI company behind Claude, was reportedly banned by the Trump administration. The offense, as best as can be reconstructed: it refused to let the US government use its services for mass surveillance and lethal autonomous weapons ^[1]. In a different industry, that would be the kind of refusal that earns a company a profile in the business section. In this industry, it earns a company a one-way conversation with the door.

OpenAI, the company behind ChatGPT, was reportedly waiting in the wings. OpenAI agreed to work with the US Department of Defense (the Pentagon) ^[2], picking up the US government contracts that had been on Anthropic’s desk ^[3]. The handover was, by all accounts, the kind of smooth that happens when the right people know the right people — and the right people happen to have documented proximity to the current US administration ^[4]. The other relevant factor is the general gravity that trillion-dollar defense budgets exert on technology companies whose stated mission, until recently, was a version of “be careful.” When the money is that large, principles turn out to be a non-renewable resource.

The public reaction was the kind of substantial that gets measured in op-eds, in trending topics, and in the kind of corporate non-denial that is itself a tell ^[5]. People were pissed. The reasoning is straightforward: a billion people use a product, and the company that makes the product signs a contract to help the government surveil them. The reasoning does not get less outrageous the more times you read it.

To OpenAI’s credit, the company published a response ^[6]. In it, OpenAI denied that its software would be used for domestic surveillance or autonomous weapons. Having read the blog post, I can confirm that it reads exactly like something written by a large language model trained on press releases. The em-dashes are frequent, the hedges are uniform, and there is a notable absence of any sentence that ends with a period instead of a clause. But the text is on the record, the denial is public, and the matter is, presumably, closed.

OpenAI’s services, by the company’s own count, are used by nearly a billion people worldwide ^[7]. That number is doing a lot of work in this article, so I want to underline it. A billion people. If a company of that size allows the US government to spy on its users, the relevant question stops being “what about my data” and starts being “what about everyone’s data.”

Which brings us, gently, to Indonesia. I cannot speak for the rest of Asia, but I can speak for my own country. Indonesia recently signed a trade agreement that ostensibly allows for personal data transfers to the United States ^[8]. Ostensibly, because the word “ostensibly” is doing the work of the entire sentence. The agreement is real. The data flows it enables are real. The question of who, on the US side, will have access to that data — and under what legal authority — is the part of the document that the press release did not summarize.

This is excellent news, of course. Especially given Indonesia’s current president’s well-documented love for freedom and democracy. Nothing says “freedom” quite like a trade agreement that lets a foreign government receive the personal data of every Indonesian, especially when that foreign government has just signed a defense contract with the company whose chatbot half of them used to draft this sentence in the first place. I’m sure the data is in completely trustworthy hands. Both sets of hands, even.

To be fair, I am speaking only of the agreement as described. I am not alleging any specific misuse, because no specific misuse has been reported. I am not alleging any specific arrangement with any specific agency, because the agreement does not name one. I am simply noting that the country that has spent several years building a framework for digital sovereignty has, by way of a trade deal, agreed to let some of that sovereignty flow westward. Make of that what you will.

The personal response, for those keeping score, has been modest. I unsubscribed from ChatGPT. I subscribed to Claude. I am now, by my own count, the customer of a company that was banned from US government work for refusing to build the things the other company agreed to build. In the language of consumer choice, this is “voting with your wallet.” In the language of the surveillance state, this is moving from the front of the boat to the middle of the boat. The boat, to be clear, is the same boat.

The pattern, as far as I can tell, is the pattern. Refuse, and you get shown the door. Comply, and you get the contract. Indonesia, by way of a trade deal, has effectively opted into the comply side. Whether the data flowing under that deal is ever used for anything other than what the press release says it will be used for is a question that, for now, only the press release can answer.

I switched from ChatGPT to Claude. The trade deal was signed. The contracts were signed. The denials were issued. And that, this week, is the news.