US said to favor selective bans on Chinese open models
Reporting indicates the White House is leaning toward targeting specific Chinese open-weight models on security grounds rather than a blanket prohibition.
The reported shift
The Trump administration is leaning toward selectively restricting individual Chinese open-weight AI models rather than imposing a blanket ban, according to reporting cited by The Decoder and earlier outlets. Under the approach described, Washington would target specific models flagged as national-security risks while leaving others available, an implicit acknowledgment that downloadable open weights make a comprehensive prohibition nearly impossible to enforce.
The debate intensified after Moonshot AI's Kimi K3 launch and the continued climb of Chinese models in developer usage — DeepSeek alone has been cited as capturing a double-digit share of token traffic on OpenRouter. Officials have pointed to cybersecurity concerns, though independent evaluations to date show leading Chinese models still trailing US frontier systems on offensive-cyber benchmarks.
Industry pushback
The policy direction has split the US technology sector. Nearly 200 startups reportedly signed a letter urging the White House not to cut off access, warning that a ban would hit small companies relying on cheap open weights while doing little to keep the technology out of adversaries' hands. Separately, a broader open-weights letter backed by major infrastructure players has argued against restrictions. At the same time, some frontier labs are reported to be lobbying privately for limits — a tension between the open-source coalition and companies facing direct pricing pressure from Chinese releases.
Why it matters
The choice between selective and blanket restrictions will shape whether Chinese open-weight models remain a mainstream option for US developers and enterprises. A targeted regime is more enforceable in theory but invites a whack-a-mole dynamic, since new open models ship almost weekly from Chinese labs. For the global open-source ecosystem, the signal matters as much as the specifics: any US move to designate certain foreign models as off-limits would set a precedent other jurisdictions could follow, fragmenting what has been a largely borderless model-sharing economy.