Who independently measures the dangers of AI?

Perspective Researchers, companies and regulators see it differently

From the researchers' standpoint, the point of contention is how to set the probabilities. Marcus sees human extinction as nearly zero. I think there's almost zero chance of literally extinguishing the species, which is what people talk about.── Gary Marcus (2026-10-15, source) Yet he says the probability of catastrophe is high. So the chance of catastrophe, I think is pretty high. And dystopia, I think we're already basically here.── Gary Marcus (2026-10-15, source)

Hinton says he cannot give a number but that the risk is not zero. There is no way to estimate that number,   but it's certainly not zero.── Geoffrey Hinton (2026-10-14, source) The two point in opposite directions on their extinction estimates. But they are close on the view that resources should be devoted to countermeasures.

Huang, speaking from the semiconductor industry, says companies should cooperate with one another because the technology is powerful and can be dangerous. We absolutely must work with each other, and the reason for that is because we think that this technology is going to be powerful and potentially dangerous.── Jensen Huang (2026-10-15, source) 〔Council inference〕This call for cooperation tries to secure safety through a different route from regulators issuing subpoenas.

Thought The risk-stance and architecture axes have shifted

What moved was mainly the risk-stance axis and the architecture axis. Marcus calls betting everything on a black box whose contents no one understands a fundamental mistake. what we have gone all in on is black boxes that nobody understands internally. And I think that was a fundamental mistake.── Gary Marcus (2026-10-15, source)

Hinton, by contrast, says the prediction made 50 years ago that neural networks would replace symbolic AI has proved correct. I predicted 50 years ago that neural  networks were going to take over from symbolic AI,   and I was right.── Geoffrey Hinton (2026-10-14, source) One side treats the same history of the technology as a lesson, the other as a track record.

On the human-centered axis, Hinton pointed out that few people are working on research into AI that values humans more than itself. we would like that thing to care more about   us than it does about itself. And people are not  working on that.── Geoffrey Hinton (2026-10-14, source) 〔Council inference〕This point is a blind spot for the council in the safety and alignment section.

Ideology Safety versus competition, universal rules versus national standards

On regulation, those who want universal safety rules and those who are wary of regulation split. Marcus says the US and China should agree on safety rules. I think the rules around safety should be universal. And I really think the U.S. and China should make a deal around that.── Gary Marcus (2026-10-15, source) Amodei also says it cannot be managed by a single leader, company or country. No leader, no company, and no nation can manage this alone.── Dario Amodei (2026-10-13, source)

Huang points in a different direction. His view is that building on US technology stacks and standards makes it easier to deploy worldwide. if you're an American company and you build on an American tech stack using American standards, you're fairly deployable throughout the world.── Jensen Huang (2026-10-13, source) Common rules and the spread of national standards are two different ways of picturing the same international order.

〔Council inference〕On the regulation axis, Andrew Ng's cautious position sits alongside the strong regulation arguments of Russell, Bengio and Gebru. Even where they agree on direction, they split over how binding the rules should be. Agreement is broadest on countering misuse, and Amodei cited a global ban on using AI for biological weapons. First, a global ban on using [music] AI to make biological weapons.── Dario Amodei (2026-10-13, source)

Now Rising incidents and doubts about jobs and profitability

As a fact of this round, Marcus said he had foreseen that combining coding agents would become a security nightmare, and that it came true. I said that large language models plus, uh, coating agents was going to be a security nightmare. And that's exactly what happened.── Gary Marcus (2026-10-15, source) In the site section, a report that OpenAI's AI agent spent funds from a crypto wallet without permission appears alongside a report of the FTC subpoena.

On labor, Hinton revised his estimate of when replacement would occur from five years to about fifteen. I still believe they'll be replaced, but I was   wrong about the timing. I said it would happen in  5 years. It's probably going to be about 15 years.── Geoffrey Hinton (2026-10-14, source) Marcus says junior developers are not getting hired, basically junior developers aren't getting any jobs. The senior ones will eventually age out. We may actually be in trouble.── Gary Marcus (2026-10-15, source) and that an economic downturn may also be near. we might actually get an economic crash very soon. It may have started yesterday, right?── Gary Marcus (2026-10-15, source)

Huang, on the other hand, mentioned additional hiring in the US. And, that building will hold 4,000 people. So we'll hire another 3,000 people.── Jensen Huang (2026-10-13, source) 〔Council inference〕Only two people have made their position on labor clear. The shift in the assessment reflects how few spoke more than any change of opinion.

Future Whoever carries out verification sets the direction

〔Council inference〕The question is whether the direction will move from company self-reporting to independent verification, and then to safety rules that span countries. Marcus says companies heading toward IPOs cannot be trusted. In fact, the current one is a little bit dangerous. The next model might be more dangerous, and we can't trust companies that need to IPO── Gary Marcus (2026-10-15, source) Whether a system of independent evaluation takes shape will be the dividing line.

There are three benefits. First, the additional US hiring and expanded manufacturing base that Huang describes. Second, the stance of safety cooperation among companies and improved efficiency. Third, that hybrid systems are practical enough in domains where results can be verified.

There are also three risks. First, fewer job opportunities for junior developers and a break in the passing down of skills. Second, a growing number of security incidents tied to deploying agents whose reliability remains unresolved. Third, doubts about profitability and fears of an economic downturn.

As a minority view, there is Andrew Ng's position. He holds that international safety regulation is an excessive constraint that could hinder the development and spread of the AI industry, and that competition and talent development should take priority over stronger regulation.