Open-Weights AI Policy Fight Is Starting in Earnest

Anthropic published a position on open-weights models, and the HN thread is immediately skeptical. Commenters read it as a calculated move: Anthropic endorses open-weights models that 'don't have dangerous capabilities', which is a framing that conveniently lets them support regulation that would freeze competition at the current capability level. The 'F you, I got mine' comment captures the mood of the thread.

The pattern here is that every major AI lab is now trying to use policy as a competitive weapon. The labs at the frontier have strong incentives to define 'safe' in ways that exclude models that might catch up to them. Open-source developers and smaller players are watching this closely because the regulatory definitions being set now will determine who can operate in two to three years.

The distillation ban discussion in the thread adds texture: if you can train a smaller model by learning from a larger one, the capability gap closes fast. Banning distillation is a direct attack on the mechanism by which open models compete.