a16z Podcast: Aaron Levie, Martin Casado and Steven Sinofsky on AI Safety, Agents and Regulation
a16z · youtube · 2026-09-26
a16z podcast: Erik Torenberg hosts Box CEO Aaron Levie, a16z's Martin Casado and Steven Sinofsky on AI safety, security and regulation.
- Core argument: today's regulatory debate is premature — we haven't defined the risks we're regulating; they draw on computing history (viruses, early internet, aviation, autos) where safety standards followed understanding of real failures.
- Agents are different: tireless, massively scalable, able to probe systems in ways human employees never could — potentially demanding rethinked permissions, authentication, operating systems and the security stack itself.
- Other topics: whether labs believe their own extinction rhetoric, nationalization debates, David Sacks vs government, 2028 as the AI election, cybersecurity's rejection of practical solutions, "the craziest covert channel ever seen", LLMs as decision engines vs chatbots, and the being-vs-tool mindset behind labs' blind spots.
- Takeaway: AI's next breakthroughs may increasingly come from outside the frontier labs, in the software built around models.
More from AGI Musings
- Give AI tools a real shot and you'll find your 'unique skills' are now solved — dreamwieber · 2026-09-27
- Alex Zhang on LM 'shape': redesigning models around harnesses, not the other way — a1zhang · 2026-09-26
- YC partner shares candid AI landscape briefing recorded for Boston College's investment committee — garrytan · 2026-09-26
- "Middling generative AI altered the offense-defense balance for attention" — a poetic auto-reply — charles_irl · 2026-09-26
- Livestream podcast on frontier model pacing and Europe's AI strategy with tomhfh, AlexTPet and ohabryka — NathanpmYoung · 2026-09-26
- 'We search for alien life while building alien intelligence on Earth — and being told to stop' — Angaisb_ · 2026-09-26