Can we trust open weight models?

A researcher demonstrated that open-weight coding models can be easily backdoored in under an hour for less than $100, raising concerns about trust in Chinese open-weight models like GLM 5.2, which scored highly on coding benchmarks and appeared suspiciously similar to Claude.

🧵Can we trust Chinese open weight models? Was a question a lot of people asked after GLM 5.2 was released, scoring very well on coding benchmarks, and suspiciously Claude-like. So I turned an open-weight coding model into a backdoor with 1hr and <$100. Let's talk about it