The Trump administration is leaning toward selective bans on specific Chinese open-weight AI models rather than a sweeping prohibition, according to a New York Times report. National security officials are concerned that models like Kimi K3 could be exploited, though the Times notes those fears remain theoretical for now — even recent Chinese releases trail Western counterparts by a wide margin on cybersecurity benchmarks.
Against this backdrop, OpenAI and Google DeepMind have signed an open letter opposing regulation of open-weight AI models. The list of backers reads like a who’s-who of companies with skin in the open-model game. Google DeepMind fields its own Gemma series. Microsoft and Google both sell access to Chinese models through their cloud services. And several signatories would be perfectly happy to see Anthropic and OpenAI lose some market grip. That alignment of convenience makes the letter’s principled stance a bit harder to take at face value.
What complicates the picture further is that OpenAI signed the letter while, according to the Times, privately lobbying alongside Anthropic to restrict Chinese open models. The contradiction isn’t subtle. OpenAI faces real pricing pressure from cheaper Chinese alternatives and wants the competitive threat contained — even as it publicly defends the principle of open-weight freedom. Anthropic, which doesn’t sell open models, has no such conflict. Its lobbying position is cleaner, if less surprising.
The administration’s targeted approach suggests someone in Washington understands that a blanket ban would be both impractical and counterproductive. Plenty of legitimate research depends on open-weight access. But selective restrictions raise their own questions. Who decides which models make the cut? On what criteria? And how do you enforce a ban on weights that are, by definition, already out there? Those are the details that will determine whether this is a serious security framework or just political theater.
