Definition
An open-weight model is one where the underlying trained parameters ('weights') are published publicly, so anyone can download, run, and modify it — unlike a 'closed-weight' model that only a company operates behind an API. National safety institutes now formally test open-weight models for how easily their safety restrictions can be removed and for cyber-offensive capability.
Why it matters
Executives evaluating vendor or foundation-model choices need to understand that open-weight models carry a different risk profile: once released, a company loses the ability to patch, monitor, or revoke misuse.