Rules are flexible. The people who pay attention know this.

People expect the US to apply rules evenly to both American AI efforts and that of its primary geopolitical rival. China certainly doesn't do that. Their entire economy is based around giving Chinese firms the advantage regardless of what it means for the wallets of consumers at home or abroad.

Since that's who we're playing against, and no one is going to willingly give up their open-weight models from the totalitarian rival, well, then use their ruleset.

It's that or punish the theft by making the US companies pay for their training data and blocking outside efforts that trained off the data the US companies stole.