Frontier models are uniquely dangerous, open models are especially dangerous, and therefore only a handful of enormous companies should be allowed to build them.
Then a competitor ships and yesterday’s forbidden capability becomes today’s product category.
Open models are now approaching the performance of the best closed models. This does not prove that AI carries no risk. But it proves that no frontier lab owns the frontier for long, and that policy should not turn a temporary technical lead into a permanent regulatory moat.
Open models let competitors attack the price and performance of closed APIs. But most crucially, they apply pressure on the frontier. If incumbent labs would rather lobby (sometimes so effectively they shoot themselves in the foot!) politicians than advance, an open ecosystem can advance without them.