“Tech companies like OpenAI currently make decisions about AI values largely unilaterally, with employees writing content policies that are then trained into the models.”