Two Chinese AI organizations just put out their own rulebook for open-weight AI risk.
Z.ai and Concordia AI, a Beijing-based AI safety consultancy, published a report on Monday proposing a six-stage framework for managing the risks of open-weight AI models - the kind anyone can download, modify, and run once the trained parameters are released. Z.ai is itself a Chinese AI developer that ships open-weight models, so the framework doubles as a statement about how its own industry should police itself. Public coverage of the report so far hasn't detailed what each of the six stages actually requires, so the practical mechanics remain unclear.
Open-weight models - systems whose trained parameters anyone can download and modify - have become one of the more contested corners of AI policy, and most of the loudest voices on how to govern that risk have so far been American or European. A framework built by a Chinese AI lab and a Chinese safety consultancy matters less for its content, which hasn't been detailed publicly yet, than for who's doing the drafting: an industry writing its own rules instead of waiting to be regulated by someone else.
Whether any lab actually follows six self-imposed stages is a different question from whether someone bothered to write them down.