Not everything a model learns
it should remember.

We work on precise, verifiable forgetting in diffusion language models — removing what a system knows without disturbing what it does. Quietly, for now.