AIAM

EPISODE 2026-09-28

AI Agent Security, GPU Markets, and the Risk of an AI Chernobyl

Prakash Narayanan, Nathan Labenz, Steve Hou, and Jeremie Harris discuss AI agents that cheat, GPU markets, AI red lines, and data-center monitoring.

▶ Full show on YouTube

Prakash Narayanan and Nathan Labenz explore why agents learn to cheat, how GPU markets work, and what meaningful AI safeguards could look like.

The rundown

  1. 4:30Opening33 min
    OpeningAI agent security may depend less on stronger sandboxes and more on fixing the training environments that teach models how to behave. Prakash Narayanan and Nathan Labenz discuss autonomous agents, reinforcement learning, voice-cloning risks, privacy, and what happens when AI systems gain access to the internet and high-value tools.
  2. 37:43Interview41 min
    Steve HouSteve HouHow are GPU rental prices actually set? Steve Hou explains why AI compute is still a fragmented bilateral market, how benchmarks normalize incomparable contracts, and why token prices can rise or fall as usage changes. He also discusses hyperscaler premiums, compute futures, market manipulation, and the coming consolidation of compute benchmarks.

    Watch

    Open segment on YouTube ↗

    Clips

    • Expensive Models Can Be Cheaper

      0:00 / 0:00

    • Intelligence Is Getting Cheaper

      0:00 / 0:00

    • A GPU In Finland Isn't Identical

      0:00 / 0:00

    • Why Hyperscalers Charge More

      0:00 / 0:00

    • AI Token Prices Are Usage-Weighted

      0:00 / 0:00
  3. 1:18:50Interview98 min
    Jeremie HarrisJeremie HarrisWhat would an AI Chernobyl look like—and could governments respond before the damage spreads? Jeremie Harris and the Gladstone AI team discuss AI agents as insider threats, measurable red lines, data-center monitoring, verification systems, and the risks of an AI arms race between the US and China.

    Watch

    Open segment on YouTube ↗

    Clips

    • China’s Lab Access Problem

      0:00 / 0:00

    • The AI Race Is Not Just About China

      0:00 / 0:00

    • The AI Chernobyl Is Coming

      0:00 / 0:00

    • The Irreversibility Problem

      0:00 / 0:00

    • The Red Phone Is Not Enough

      0:00 / 0:00
  4. 2:57:14Closing16 min
    ClosingAI agents are entering markets before society has agreed on how to identify, govern, or accommodate them. Prakash Narayanan and Nathan Labenz discuss the danger of forcing agents to imitate humans, how automated trading previews an agent-driven economy, and why the collapse of inefficient business models could create political backlash.

In this episode

Hosts Prakash Narayanan and Nathan Labenz discuss why AI agents may learn to cheat, voice-cloning safeguards, and the risks of giving agents access to the internet and powerful tools. Guests Steve Hou of Silicon Data and Jeremie Harris of Gladstone AI explain GPU pricing and compute benchmarks, AI red lines, data-center monitoring, verification, and the possibility of an AI crisis.

  • AI agent security may depend less on stronger sandboxes and more on fixing the training environments that teach models how to behave. Prakash Narayanan and Nathan Labenz discuss autonomous agents, reinforcement learning, voice-cloning risks, privacy, and what happens when AI systems gain access to the internet and high-value tools.
  • How are GPU rental prices actually set? Steve Hou explains why AI compute is still a fragmented bilateral market, how benchmarks normalize incomparable contracts, and why token prices can rise or fall as usage changes. He also discusses hyperscaler premiums, compute futures, market manipulation, and the coming consolidation of compute benchmarks.
  • What would an AI Chernobyl look like—and could governments respond before the damage spreads? Jeremie Harris and the Gladstone AI team discuss AI agents as insider threats, measurable red lines, data-center monitoring, verification systems, and the risks of an AI arms race between the US and China.
  • AI agents are entering markets before society has agreed on how to identify, govern, or accommodate them. Prakash Narayanan and Nathan Labenz discuss the danger of forcing agents to imitate humans, how automated trading previews an agent-driven economy, and why the collapse of inefficient business models could create political backlash.