While we are doing lobbying for AI laws, I propose this. All providers of closed source models are 100% liable for their outputs. Copyright, libel, conspiracy, fraud, etc... If you want safe harbor from this liability, release the weights. This is real AI safety legislation.