Rumor going around Chinese AI circles today: future Chinese open-weight models… · @Moblacksoil · TwKit

@Moblacksoil

Rumor going around Chinese AI circles today: future Chinese open-weight models may ship with offensive cyber capabilities stripped out before release. Think pentesting help, exploit writing, red-team work. The source is an unverified group-chat screenshot about a regulator meeting. Nothing official yet. If it's real, two things follow: 1. Older open weights that are already out stay as they are. Expect people to start archiving them. 2. Security benchmarks won't be comparable across versions anymore, so a lower score may not mean a worse model. Security researchers: would this push you to closed APIs or older local weights?

曝光 252 · 评论 0 · 点赞 3 · 书签 0