An observation of the Anthropic | OpenAI | DoW discussion is that many seem to think of a commercial contract like they think of AI alignment. A binding commitment that would prevent anyone from doing anything wrong with it after. It's wrong about alignment and it's wrong about contracts.

In reality, like AI, contracts are messy things. You get the best safeguards you can, enough that you can trust the others to act appropriately, and decide to work together. There is no contract on earth that will bind the government, or indeed any other counterparty, to adhere to your particular standards forever. That's simply not a thing that exists.

Usually you can see this when the lawyers start throwing worst case scenarios at you. At some point you have to stop and say, if we really are going to fight over those points today, there's not enough trust that we should do the deal in the first place.