We often struggle in AI governance and ethics with continuous scales: When is a model fair enough to launch? When is it safe enough? At what point is a model too risky to be released in a certain way?

From what I've seen, the challenge is that without clear expectations, no position ever feels stable; every incremental move can be considered responsible. I believe that one of the greatest opportunities for AI governance is to provide principles with clear expectations to align on.

In negotiation, if someone says a price is $100 and "no dollar less," that sounds much more credible to me than if that same person had said, "Okay, I'll do $97 but no dollar less." By moving just three dollars, we move from a stable commitment to an unstable point where another concession just reduces the price but doesn't affect any principle.

What we need are principles strong enough to serve as resting points for clear expectations, but flexible enough to be revised over time as we learn.