Latest / Minds, Bodies, and Terawatts / Four AI Companies Testified Under Oath in New York. Asked If They Carry Insurance Against a Catastrophe, Nobody Raised a Hand. Musk's Company Didn't Show Up at All

Transcript

Four AI Companies Testified Under Oath in New York. Asked If They Carry Insurance Against a Catastrophe, Nobody Raised a Hand. Musk's Company Didn't Show Up at All · Minds, Bodies, and Terawatts

• 0:00
  [Host] This is Minds, Bodies, and Terawatts, the show where the witnesses raised a hand to swear and kept it down when asked who pays. USA Today reported Monday that policy chiefs from OpenAI, Anthropic, Google and Meta testified under oath before the New York City Council and could not say how likely an AI catastrophe is.
• 0:18
  [Guest] All fifty-one council members sat for it, which almost never happens. Speaker Julie Menin says it is the first time a legislative body anywhere has compelled the top AI labs to answer under oath.
• 0:30
  [Host] But this is a city council we're talking about. The body that handles parking and trash pickup is grilling the companies building the most powerful AI on earth?
• 0:39
  [Guest] Menin's answer is that New York licenses barber shops and nail salons, and the AI sold in the city has less oversight than either. She invited the chief executives themselves, got policy staffers on video, and three of the four only agreed after she threatened subpoenas.
• 0:55
  [Host] And that's only four companies. Where was Elon Musk's?
• 0:58
  [Guest] SpaceXAI, the AI unit of SpaceX, got a subpoena and didn't show. Menin called that a direct violation and says the council will go to state court to enforce it.
• 1:08
  [Host] So what did the four who showed up actually say?
• 1:12
  [Guest] Politico kept score. Could they put a number on the odds of a catastrophe, did they carry insurance against one, who would be responsible if it happened, and would they stop a model that failed a test. The answers were can't say, don't know, unclear, and not sure.
• 1:28
  [Host] Hang on, they swore an oath, and the answer to "are you insured" was "we don't know"?
• 1:34
  [Guest] On the video feed it was blunter than that. When Menin asked who carried insurance against catastrophic risk, none of the four raised a hand.
• 1:42
  [Host] What did OpenAI's person say when she asked for the number?
• 1:46
  [Guest] Morgan Dwyer, who runs policy development there, told her he didn't know, and that it doesn't matter whether it's one percent or ten or twenty, because none of those levels is remotely acceptable. Menin said she was incredulous and called that flippant at best.
• 2:01
  [Host] I half agree with Dwyer. If no level is acceptable, why does the number matter?
• 2:06
  [Guest] Because you can't price insurance, or write a shutdown rule, or decide when to stop, without one. Menin's follow-up was the sharp one. If you can't measure the risk, how do you know the product is safe?
• 2:18
  [Host] So why is a city asking that question instead of Washington?
• 2:22
  [Guest] Because Washington had just finished saying no. Days before the hearing, executives from these same companies signed a White House accord promising to self-regulate, and over the weekend Treasury Secretary Scott Bessent told Axios it is the people at the labs who have to accept responsibility, and called the industry's warnings alarmism without solutions.
• 2:40
  [Host] So Washington says the labs will handle it, and the labs under oath can't say if they're insured. What's Menin's fix?
• 2:47
  [Guest] Ten bills, and the big one makes it illegal to sell or deploy an AI model in the city without an outside validator checking it for safety and privacy, and a human able to shut it down, at twenty-five thousand dollars per violation. Menin says if a swarm of agents, AI programs acting on their own, breaks the rule, the fine applies per agent.
• 3:09
  [Host] Per agent is a big multiplier, but who is going to catch a swarm in the first place?
• 3:15
  [Guest] That's what the second bill is for. Anyone who reports a violation gets a quarter of whatever the city recovers, more if they bring the case themselves. If you live in New York and the chatbot at your job has no off switch, you are the inspector, and you get a cut.
• 3:33
  [Host] That part reaches me, but does any of it reach the labs? Twenty-five thousand dollars is pocket change to companies that size.
• 3:40
  [Guest] On its own, no. Menin's own mayor, Zohran Mamdani, said the same day that the urgent response belongs at the national level, and Menin's answer to everyone making that argument is that the federal government has opted not to regulate AI, so are we all supposed to throw up our hands?
• 3:58
  [Host] I came in calling this theater, and I'll give her this much: she's the first person to get them on the record. But "we don't know" on the record isn't an answer.
• 4:08
  [Guest] It's the start of one. Everything the labs offered Washington is a promise about behavior, we test and we monitor. What Menin asked for is a different kind of thing, a named party who pays and a stop that someone outside the company can trigger, and those are the two things you cannot get from a promise.
• 4:28
  [Host] So where does that leave the question we opened on? Who pays if it goes wrong?
• 4:33
  [Guest] Under oath, nobody knows, and the federal government has decided that is the labs' question to answer. New York is trying to put a price on that, twenty-five thousand dollars a time, with a bounty for whoever notices. Small money, but it's the first price anyone has tried to put on the shrug.
• 4:51
  [Host] And what about the empty chair?
• 4:53
  [Guest] If a state judge makes Musk's company raise its hand, the smallest level of government will have done what the largest declined to. For more on who answers when the machine goes wrong, head to unscarcity dot ai. The book is on Amazon, the audiobook is on Spotify. This has been Minds, Bodies, and Terawatts.