Sriram Krishnan says open-weight models are easier to secure because anyone can inspect them

pstAsiatech · x · 2026-07-21

Sriram Krishnan argues that open-weight models are easier to trust for code defense because anyone can download them, inspect them, fine-tune them, and analyze them directly. He says he wishes leading open models were American, but the broader security point is that openness changes the threat model: like Linus Torvalds’ “given enough eyes, all bugs are shallow,” open weights let the whole world look inside, unlike closed models. The quote also name-checks Moonshot, DeepSeek, and Qwen as examples of strong non-American innovation.

Related event: Sriram Krishnan: Open-Weight Models Are Safer(2 posts)→

Original post →

More from Safety

Safety channel →