Aaron Levie, Sinofsky & Casado: Don't Regulate AI Agents Before Defining Risks
a16z Podcast · rss · 2026-09-26
On the a16z podcast, Erik Torenberg hosts Box CEO Aaron Levie, a16z's Martin Casado, and Steven Sinofsky to debate safety, security, and regulation for AI agents.
Key points:
- Today's regulatory conversation is running ahead of a clear definition of the risks being regulated
- Lessons from computer viruses, the early internet, aviation, and autos: safety standards work best built after understanding how technologies actually fail
- Agents differ from human employees — tireless, operating at massive scale, probing systems in ways humans never could — which may demand rethinking permissions, authentication, operating systems, and the entire security stack
- AI innovation may increasingly move beyond frontier labs into the software built around models
More from AGI Musings
- Joke Paper Title 'Can a Computer Scientist Align an LLM?' Riffs on Classic Essay — MichaelRetchin · 2026-09-26
- Nobody Rations a PDF: Can AI Push Physical Goods Into Abundance? — r0ck3t23 · 2026-09-26
- Neural networks don't literally mirror the brain: the name descends from a 1943 math abstraction — vishalmisra · 2026-09-26
- The Cognitive Revolution debuts AI-co-written novel about benevolent superintelligence — The Cognitive Revolution · 2026-09-26
- mitsuhiko: Don't dunk on local model tinkerers — they enable experimentation you can't get elsewhere — mitsuhiko · 2026-09-26
- Neal Mohan Says YouTube Killed the Gatekeeper — It Just Moved Into Gemini — cen6wkf · 2026-09-26