Trump Admin Weighs US Open-Model Cap Tied to Chinese Peers

A quiet Washington conversation about open-source AI is worth paying attention to. The Washington Post reported on July 13 that the Trump administration and AI industry groups are discussing a framework that would streamline the release of US open-source AI models, with a capability ceiling set at where the leading Chinese open models sit today.

The logic behind the ceiling is that Chinese models reportedly trail US frontier models by an average of seven months. A US lab releasing open weights at or below the current Chinese frontier would, under this framing, not be giving anything away that an adversary cannot already download. The industry side of the conversation expects Chinese 'Mythos-class' capabilities to be available for free download on the internet in six to 12 months, which reframes the argument from whether the US should release open weights at all to how much longer it can defensibly not.

The backdrop is fresh and painful for one specific company. On June 12 the Commerce Department abruptly ordered Anthropic to deny access to its Mythos 5 and Fable 5 models for any non-US persons, and Anthropic pulled both worldwide before international access to Fable was later restored with restrictive guardrails. Chinese labs used the window to push their own open-weight releases. Andreessen Horowitz partner Martin Casado said on the record that when a startup walks in pitching an open-source stack, there is roughly an 80% chance it is running on Chinese weights, though he later clarified that share is of the subset of startups using open source at all, not of every startup.

The honest caveat is that this is a single Washington Post scoop about talks, not a signed order. There is no named agency owner in the reporting, no draft text, no public list of participating companies, and no answer yet to the operational question of who measures 'capability' and how often the ceiling moves as Chinese labs ship. Take the six to 12 month Mythos-class timeline as an industry projection, not a settled forecast.

If the framework lands, the winners are the US labs that had been holding back open-weight releases pending regulatory clarity, and the investors who have been reluctant to underwrite startups building on Chinese stacks. The more interesting political test is whether it also gives the administration a way to answer critics of the Anthropic action without formally reversing it.