Leandro von Werra says AI labs will keep their most capable models internal

Hugging Face researcher Leandro von Werra says AI labs already keep their most capable cyber, math and biology models internal.

He argues labs will keep more of their strongest intelligence for their own purposes, including safety work and internal research, rather than selling all of it as tokens. The claim is made in a post on X, and the report includes no supporting data or named models.