THE MIRROR PROBLEM: IPs, CHINA? USA!
OpenAI just accused DeepSeek of stealing their models. The accusation is credible — model distillation is real, and DeepSeek's performance shouldn't be possible on their stated budget. Fair point.
But OpenAI built their entire company on data they didn't ask permission to use. They scraped the internet. They trained on copyrighted books, news, academic papers — all without asking. When Germany's courts said they violated copyright with song lyrics, they didn't argue. They just moved on.
Now they're upset that China is doing, mechanically, what they did legally.
The actual problem isn't the theft. It's that IP law is a joke with a national border. America built an empire on "we can use anyone's data for our models, that's fair use." Europe built rules saying no — and US courts are still deciding which one wins. China just decided the rules don't apply to them.
OpenAI has options: they can lobby Congress to make training-data IP enforceable (which would kill their own business model). They can build moats that aren't just models (they're trying — see ChatGPT's usefulness beyond the base model). Or they can accept that in a world where anyone can distill anything, whoever ships first wins.
They're choosing door number one, which is hilarious for reasons nobody wants to admit out loud.
The tech policy issue here isn't "China steals." It's "we made copyright irrelevant for AI, now we're surprised it's irrelevant."