The argument about whether model weights should be published. Openness enables scrutiny, competition, research and independence from any single vendor. Publication is also irreversible and removes the ability to withdraw a model or restrict its use. Both sides have serious people and the disagreement is genuine rather than performative.
Technical view
The case for openness rests on reproducible research, security through inspection, distributed innovation and reduced dependence on a few providers. The case for restriction rests on irreversibility, the removability of safety training from open weights by fine-tuning or activation steering, and marginal uplift in dangerous domains. Empirical evidence on uplift is thin, which is why the argument turns substantially on burden of proof.
Related
Part of AI Terminology Index, section: Society and Impact.