Open-Weight AI Wins a Conditional Policy Truce

The argument over open-weight AI models, framed for years as innovation versus catastrophe, has settled into a conditional truce. Major jurisdictions have declined to restrict releases broadly, opting instead for transparency obligations, evaluation duties at defined capability thresholds, and reserved authority they describe as break-glass rather than default.

The economic evidence carried the argument. Open models became the substrate of startup ecosystems and national AI strategies alike, and restricting them came to look like restricting one's own industrial base while rivals published freely.

The truce's conditionality is the point: thresholds will be contested, evaluations will mature, and a sufficiently alarming capability could reopen everything. For now, policy has landed where the technology community largely predicted it would, watchful permission, with the paperwork to prove the watching.