Latest / Minds, Bodies, and Terawatts / The One AI Regulator to Reach Trump's Desk Was Designed by Google's Hassabis, and Zuckerberg, Musk and Nvidia's Huang Each Called the President to Kill It

Transcript

The One AI Regulator to Reach Trump's Desk Was Designed by Google's Hassabis, and Zuckerberg, Musk and Nvidia's Huang Each Called the President to Kill It · Minds, Bodies, and Terawatts

• 0:00
  [Host] This is Minds, Bodies, and Terawatts, the show where the seating chart is the policy. The Wall Street Journal reported on September 17th that Mark Zuckerberg, Elon Musk and Nvidia's Jensen Huang each spoke privately with President Trump to argue against an AI regulator, and the President dropped it.
• 0:17
  [Guest] Three conversations, three billionaires, one proposal. Google's Demis Hassabis had drawn it up: a body paid for by the industry that would test the most powerful models before the public got them, modeled on the body that polices Wall Street's brokers.
• 0:32
  [Host] We covered this in May, when Musk and Zuckerberg talked Trump out of signing an executive order on vetting AI models. What's different this time?
• 0:41
  [Guest] The thing they killed. Then it was the government's plan. This time it was the industry's own plan, written by one of the three biggest labs, and the argument against it was that it would hand those three labs the keys.
• 0:56
  [Host] Wait, Musk and Zuckerberg run AI labs. Their objection was that a regulator would be too cozy with AI labs?
• 1:04
  [Guest] With the wrong ones. The Journal says the three warned it would concentrate influence in OpenAI, Anthropic and DeepMind, and asked who would run the body and pick its members. A club with three chairs, and none of them ours.
• 1:18
  [Host] Is that a bad argument? A referee paid by the players sounds like a referee who swallows his whistle.
• 1:25
  [Guest] Critics say that about the Wall Street version too, and it's not stupid. The trouble is what replaced the club: not a better regulator, no regulator, decided by whoever reached the President last. That's a club with one chair, and it's taken.
• 1:40
  [Host] Who inside the White House lost?
• 1:43
  [Guest] The Journal says the President's own chief of staff, Susie Wiles, and Treasury Secretary Scott Bessent have been pushing for more scrutiny, and that Bessent's fear is AI-driven cyberattacks on the banking system. When that vetting order died, the two of them learned only afterward that David Sacks, who advises the President on AI, had killed it in a last-minute call.
• 2:05
  [Host] Does the White House even admit that the policy is whoever calls last?
• 2:09
  [Guest] The President says it himself. That week he posted that the only guardrail AI needs is a strong, high-IQ president, and officials have told AI executives that a deal is hard when every chief executive can dial that president directly.
• 2:24
  [Host] What did the three say in public that week?
• 2:27
  [Guest] Huang told a Salesforce conference that we don't need any new laws, that safety is an engineering problem, not a legal one. Zuckerberg posted the same day that competition and significant liability already give every lab reason enough to build safely, and pointed to Meta holding back its own agent to fix its security.
• 2:47
  [Host] And Musk? The weekend before, he posted that Anthropic's boss was right to want a slowdown.
• 2:53
  [Guest] CNBC called it a head-spinning week. He endorsed the slowdown, then told a conference that regulation is a one-way ratchet and that labs should test each other's models instead. And the Journal says he had already made his case to the President, weeks earlier.
• 3:09
  [Host] Zuckerberg says liability already does the job. Does it?
• 3:12
  [Guest] Ask what a lab is actually on the hook for. A surgeon comes with a board that can pull her license, an insurer that posted money against her mistakes, and a duty of care she never got to opt out of. A model comes with a terms-of-service page, a document written so that nobody is responsible for anything.
• 3:33
  [Host] So when he says significant liability, he means what, exactly?
• 3:37
  [Guest] A lawsuit, filed after the harm, that the courts still haven't decided how to handle. The body Hassabis drew up would have been the first thing resembling a board, one that tests before release and can say no. That's what three private conversations removed.
• 3:53
  [Host] What does that mean for me, using this stuff every day?
• 3:57
  [Guest] The Wall Street version can fine a broker or bar him from the business, which is why your retirement account has someone to complain to. For the model on your phone there's no equivalent. The test lab for it is you.
• 4:12
  [Host] Back then the question was whether the people being vetted could veto the vetting. Where does this leave it?
• 4:19
  [Guest] They still can, and now they veto each other. The vetted split into two camps, one camp designed the referee, the other camp got him fired, and the President settled it in private. We only know because the Journal found people who were in the room.
• 4:36
  [Host] And next week they're all in another room.
• 4:38
  [Guest] Next Thursday the White House hosts a state dinner for Xi Jinping, and the guest list it put out on Friday has Musk, Huang, Sam Altman, Sundar Pichai and Jeff Bezos at the table, with AI safety expected on the agenda. The men who just decided America has no referee will be the ones discussing one with China.
• 5:00
  [Host] Is there any version of this where the rest of us get a chair?
• 5:05
  [Guest] Not while the vote goes to whoever last had the President's ear. That is exactly the capture the Diversity Guard in Patrick Deglon's Unscarcity exists to catch, a standing check that spots one small group quietly taking over the reviewers and forces a reset before it hardens. For more on who gets a chair when AI gets a referee, head to unscarcity dot ai. The book is on Amazon, the audiobook is on Spotify. This has been Minds, Bodies, and Terawatts.