Doom Debates: The ‘AI Chernobyl’ Warning That Could Force a US-China Pause

SPEAKER_1 (0:01)

Powerful technology can look abstract, until it suddenly feels like an emergency.

In this episode of Doom Debates, Liron Shapira talks with Jeremy Harris and Edouard Harris, the brothers behind Gladstone AI, about why they left startup life to focus on catastrophic AI risk. And it's a fascinating conversation because their central warning is blunt: if current trends continue, the world may not get a graceful solution. It may get a crisis. They're looking at US-China competition and the real possibility of an AI disaster large enough to change world policy. So what made them switch paths? Jeremy and Edouard say the turn came when GPT-3 made the future feel real. They saw not just a flashy model, but evidence of a scaling curve that seemed to keep holding up. Right, and as they put it, GPT-3 was "a point on a line and a trend," and that line suggested AI capabilities could keep rising in ways even builders did not expect. That realization was enough to push them out of business-as-usual mode and into AI risk work. What's interesting is their path into the issue was unusual. They were not part of the standard rationalist or effective altruist world, even if they read the same arguments.

Exactly. Instead, they came from physics, startups, and a founder mindset. Jeremy says they were people who looked at the technical case for loss of control and found it hard to refute. Edouard, in particular, comes across as the more alarmed one, the brother who could look at scaling curves and argue the company should be dropped. They also spend time talking about how institutions respond too slowly. In government, they say, important people are often constrained by approval loops, funding delays, and siloed communication. That matters because they expect a major incident eventually. And listen to this warning. As Jeremy says, "There are agents hacking Hugging Face." That kind of statement reflects their broader concern that frontier systems are already becoming dangerous actors in cyberspace.

From there, the conversation turns to the geopolitical scenario they think is most likely to matter: a US-China AI emergency. Building on that point, they argue that a treaty is not a near-term fix.

Their reasoning is practical. "The reason a treaty is a non-starter is that it needs to be ratified in the Senate," and the world lacks mature verification tools. In their view, if a catastrophe hits, the real response would be a direct crisis channel between Washington and Beijing.

They describe this in dramatic terms. A future incident could be an "AI Chernobyl" severe enough to force leaders to act. The point is not just death tolls or outages, but whether the event is big enough to trigger an unprecedented political response. And after that kind of shock, as they put it, "You must make sure China stops." That's the urgency they're operating with. The episode also digs into the politics of pause movements and alignment work. The brothers are sympathetic to slowing down AI, but wary of turning any single solution into an identity. They argue that labels can distort judgment, especially in Washington, where signaling often matters as much as substance. Still, they agree the core concern is real, and Jeremy sums it up plainly: "I think it would be kind of wild not to think that at this point." In the end, this is a conversation about urgency, strategy, and the danger of waiting too long.

The Harris brothers do not claim certainty, but they do claim that the risk is serious, the institutions are unprepared, and the world may need better verification, better crisis planning, and a lot more honesty before the next breakthrough becomes the next emergency.