The threat is that governments will want specific training so that the AI is pro-democrat or pro-republican or pro-Conservative Party or pro whatever the govern

The threat is that governments will want specific training so that the AI is pro-democrat or pro-republican or pro-Conservative Party or pro whatever the government is pushing. This is the problem with foundational models. You want it to align with human values? Well, which values? Who's values?