NEW piece from me in @asteriskmgzn on the case for diffusing ai expertise and oversight beyond the executive branch. ai governance that depends on executive authority will always be more brittle. executive orders can be rescinded, administrations can be hostile or non-compliant, agency policy changes. but better policy & institutional design can make ai oversight sturdier. i use the engineering concept of “graceful degradation” — structuring systems to function even while failing, like a plane landing safely after an engine dies — as a design principle for ai policy. in practice, that could mean: statutory protections with private enforcement, technical capacity housed outside the federal executive (including in congress and among state AGs), and ensuring the executive isn’t a chokepoint in critical ai oversight. tl;dr, we should design ai governance to endure across administrations.