China's Open AI Models Are Testing America's Approach to AI Safety

The growing global popularity of open-weight artificial intelligence models developed in China is testing the assumptions underlying America's approach to AI safety, according to a report in Scientific American.

The issue comes as US President Donald Trump and Chinese President Xi Jinping prepare to discuss AI safety as part of broader talks between the two countries.

Open-weight models and safety limits

Open-weight models, which make their underlying parameters publicly available, allow developers and researchers around the world to modify and deploy the systems independently, often stripping out the safety restrictions originally built in by the model's creators.

This dynamic highlights a core limitation of an AI safety approach that relies primarily on developer-controlled safeguards: once a model's weights are released publicly, the original developer loses much of its ability to enforce restrictions on how the model is used downstream.

The rising popularity of Chinese open-weight models has intensified debate in the US over how AI safety should be regulated at a broader, systemic level, rather than relying solely on individual companies to police their own technology.

The upcoming discussions between Trump and Xi are expected to touch on AI safety alongside other technology and trade issues, though specific commitments on the topic have not yet been detailed.