a16z Podcast with Box CEO: Don't Regulate AI Before Risks Are Defined
On the a16z podcast hosted by Erik Torenberg, Box CEO Aaron Levie, a16z partner Martin Casado, and former Windows chief Steven Sinofsky discussed AI safety, regulatory pacing, the 2028 "AI election," and the Jevons paradox under the theme "We Must Pace the Frontier." Their core conclusion: regulation should not proceed before risks are clearly defined, and frontier breakthroughs may come from outside big labs.
Confirmed
- All three guests argued that current AI safety discussions lack a clear definition of risk, making regulation premature.
- Sinofsky warned AI regulation could repeat GDPR's mistakes: safety prompts on every Agent write operation would ultimately harm usability and innovation.
- Sinofsky said Jevons has made probabilistic programming the coolest area of computer science.
- Casado argued frontier labs missed Jevons because LLMs follow a text-in/text-out chat paradigm—labs are building "talking beings" rather than other product forms.
- The discussion also covered the 2028 "AI election" and the Jevons paradox.
Why it matters
- This is a systematic a16z statement on AI regulation, tied directly to Europe's GDPR lessons, with relevance for compliance design of Agent products.
- Casado's critique of labs' chat-paradigm limits suggests the next breakthrough may come from teams outside big labs, offering a lens on startup opportunities.
2026-09-26 ~ 2026-09-27 · 5 related posts
Primary sources
- [source] a16z Podcast: Aaron Levie, Martin Casado and Steven Sinofsky on AI Safety, Agents and Regulation — a16z · 2026-09-26
- [source] Box CEO Aaron Levie and a16z on AI regulation, agents and security — a16z · 2026-09-27
- Sinofsky fears 'GDPR for AI': every agent write action could trigger safety warnings — Scobleizer · 2026-09-27
- Martin Casado: labs missed Jev because 'beings speak' — software needed a model that chooses — multiply_matrix · 2026-09-27
- Sinofsky on a16z podcast: probabilistic programming is the future of AI interfaces — multiply_matrix · 2026-09-27