China has proposed new export controls on AI models, agents, and chip design technologies, signaling a tightening grip on its rapidly advancing tech sector. The draft rules, jointly issued by the ministries of commerce and information technology, restrict the export of leading AI models trained on sensitive data and advanced AI agents used in critical infrastructure, defense, and government services. The move aims to close regulatory loopholes, particularly in light of Meta’s $2bn acquisition of AI app Manus, which Chinese authorities later ordered unwound.
The proposed controls also cover technologies for designing advanced processing and memory chips, as well as packaging and networking equipment. While products built using these technologies may not be directly controlled, the rules would prevent Chinese firms from transferring underlying tech overseas via foreign acquisitions. Open-weight AI models—those publicly available and not trained on sensitive data—are exempt, unless they exceed 2 trillion parameters or rank among the top global performers.
Moonshot’s K3 model, with 2.8tn parameters, falls under this threshold. Simpler approval processes are outlined for Chinese companies building overseas data centers, manufacturing chip designs abroad, or collaborating on international standards. The commerce ministry emphasized that the controls target the transfer of restricted technology from China, not indirect access via apps or websites.
The proposal also includes controls on biomedical and solar technologies, reinforcing Beijing’s strategy to maintain security and controllability in strategic sectors.
Source: Financial Times Companies · Summarized by HeadlinesBriefing