Sam Altman tells @TheoVon that “no one was thinking” about privacy in AI before “like a year ago.” (Laughable) Claims once again that he is helpless to do anything until “the law” codifies some kind of protected “confidentiality” for AI chats. No, the problem is that companies like OpenAI have architected their products in a way that requires users to send their sensitive chats and prompts to servers controlled by those companies. That means OpenAI employees can access them—and that also means the government can access them. In June, the Southern District of New York ordered OpenAI to retain and hand over its user chat logs *regardless of promises it has made in its contracts with customers.* In other words, the state can override the terms of any private contract to violate user privacy. OpenAI can comply with this mandate because it has architected its system in a way that makes that compliance possible. It didn’t—and still doesn’t—have to be this way. But Sam knows that playing ball in the U.S. means giving law enforcement the “backdoors” they always clamor for. As the Cypherpunks knew, technologies must be built that protect individuals in an era where networked compute is dominated by an alliance between corporations and governments. There are some companies building such technologies today, but they don’t typically raise billions of dollars of venture capital or win insane government contracts.