AI regulation is being written before anyone has defined what risk it is meant to address. Aaron Levie, Martin Casado, and Steven Sinofsky argue that the current safety debate is structurally premature, driven by speculation rather than observed, measurable harm from deployed systems.
The conversation anchors itself in historical precedent. Computer viruses, the early internet, aviation, and automobiles all generated regulatory frameworks after failure modes were understood, not before. The panel uses these cases to question whether AI is being treated as a unique existential category or simply the latest iteration of a familiar problem that institutions have solved before.
The real argument worth reading for is how the group handles agentic AI specifically, systems acting autonomously in the real world, and whether the big labs are actually the right locus of innovation or safety governance going forward. The tension between incumbents and outside actors is where this discussion gets sharp.
[WATCH ON YOUTUBE →]