Latest / Minds, Bodies, and Terawatts / Trump Pulls AI Executive Order After Lobbying from Tech Giants
Transcript
- 0:00[Host] This is Minds, Bodies, and Terawatts, where the regulators show up late and leave early. Forbes reported on May 22nd that President Trump scrapped a signed and ready executive order on AI national security vetting after a single afternoon of lobbying from Elon Musk, Mark Zuckerberg, and AI czar David Sacks.
- 0:21[Guest] The order had a number, a binder, and a podium. Three phone calls later, the podium went away.
- 0:28[Host] What was actually in the order before it died?
- 0:32[Guest] A government framework to vet frontier models for national security risk. Pentagon-style red team for any AI above a compute threshold, with disclosure obligations to a federal review body before deployment.
- 0:45[Host] And the three names who killed it: Musk runs xAI, Zuckerberg runs Meta AI, Sacks is the White House AI czar. The people being vetted got to veto the vetting.
- 0:57[Guest] There's a machine-learning joke buried in that sentence. If you ask who regulates AI, the training data is "people who profit from AI." The model converges exactly where you'd expect.
- 1:09[Host] Sacks is a former Craft Ventures general partner with stakes across the AI stack. He's writing policy for companies he's invested in, and those companies are calling him to kill it.
- 1:22[Guest] Forbes places Musk lobbying Trump personally at Mar-a-Lago over the weekend. Zuckerberg's outreach ran through Meta's Washington office.
- 1:30[Host] What were the arguments?
- 1:31[Guest] Two lines. One, any vetting regime hands China a head start. Two, Biden tried this in 2023 and it slowed shipping. Both get you to no federal eyes on the model before it goes live.
- 1:46[Host] Who loses?
- 1:46[Guest] National security staff at Commerce and DOD who spent six months drafting the framework. State AGs who now have to build their own standards. The killed order would have preempted California's SB 53 on frontier model transparency and New York's RAISE Act. Now it's all live.
- 2:05[Host] There's a piece on unscarcity dot ai called Amazon's AI Penalized Women, Caesar Became Dictator. It opens with the Reuters October 2018 report by Jeffrey Dastin showing Amazon's internal hiring tool downgraded resumes containing the word "women's", caught only because engineers ran a four-year retrospective audit.
- 2:24[Guest] Deglon's argument isn't just that bias slipped through. It's that Amazon trained the tool on ten years of male-dominated resumes, so the model learned that "male" equaled "qualified." The mechanism wasn't malice. It was unsupervised pattern-matching on a tilted dataset, deployed before anyone checked.
- 2:43[Host] So the article's two load-bearing principles are transparency and power decay. Which one just failed today?
- 2:49[Guest] Transparency. Pre-deployment vetting exists so an outside body sees the model's behavior before the public meets it. Killing the order means the first audit of a frontier model is its first deployment.
- 3:03[Host] Amazon's resume tool problem at civilizational scale. Nobody knew it screened out women until they looked, and nobody looked until after it ran.
- 3:12[Guest] Right. Federal lobbying hit five point zero eight billion dollars last year per OpenSecrets, a record. The Iron Law dynamic Robert Michels described in 1911 now runs with a billion-dollar accelerator pedal.
- 3:25[Host] Sacks pushed back on the Forbes piece on X, calling it inaccurate without specifying what was wrong.
- 3:32[Guest] He didn't say the meeting didn't happen. He said the framing was off. Meanwhile three companies building the most powerful systems in history are also deciding what oversight applies to them.
- 3:45[Host] This is what Patrick Deglon calls the failure of Truth Must Be Seen in Unscarcity, when the people inside the black box are also the ones deciding whether anyone gets to look. For more on AI governance and the Iron Law of Oligarchy, head to unscarcity dot ai. The book is on Amazon, the audiobook is on Spotify. This has been Minds, Bodies, and Terawatts.