Hinton answers Huang’s AI risk criticism and rejects company self-regulation

Not knowing exactly how dangerous AI will become is no reason to leave its developers in charge of the safeguards, Geoffrey Hinton argues. In a Guardian interview published October 13, 2026, the computer scientist defended his extinction-risk warning against Nvidia CEO Jensen Huang’s criticism and rejected the idea that AI companies should regulate themselves.

A warning, not a timetable

The interview, conducted by the Guardian’s Leyland Cecco in Toronto, gives Hinton a chance to answer a pointed challenge to his warning. His estimate puts the possibility of AI causing human extinction at 10–20% in the coming decades. Hinton says the number communicates a serious possibility, rather than a precise prediction of what will happen.

Huang disputed that framing in an appearance on Ezra Klein’s podcast, quoted by the Guardian. He said Hinton’s figure was “not grounded on science” or research and called such predictions “hurtful.” Huang also argued that being an alarmist does not automatically amount to doing social good, urging a more evidence-based approach.

Hinton’s reply turns on how people respond to uncertain danger. He offered a bus analogy: someone told there was a 10% chance of a crash would not get aboard. The purpose of giving a number, he said, was to communicate the seriousness of the risk while acknowledging that its size remains uncertain.

He conceded that neither AI’s critics nor its boosters can precisely predict the technology’s future behavior. But he argued that a loss of human control would be unlike a failure society could learn from and reverse. In his account, an AI takeover would not offer a second attempt.

The point of giving numbers is to say this is a serious risk. We don’t know how serious, but it’s serious.

Geoffrey Hinton, speaking to the Guardian

Who gets to write the warnings?

Hinton’s objection to company self-regulation was blunt. He compared it with allowing oil and asbestos companies to oversee themselves, then asked whether tobacco companies should have been allowed to design cigarette-pack warnings. His imagined result was a warning printed in tiny, dark-gray lettering against black—a message technically present but difficult to read.

He also challenged President Donald Trump’s use of “superintelligence” as a replacement name for AI. Hinton said the term has a specific meaning: intelligence beyond that of people. Applying the label to AI generally, he argued, does not make a system smarter than humans. His criticism concerned the distinction between a capability and a name.

The interview continues the public role Hinton took on after leaving Google in 2023. He resigned so he could speak freely about AI’s risks, the Guardian reported. His latest remarks address both the credibility of his warning and whether the companies developing powerful systems should be trusted to set their own safeguards.

Smarter is not the whole goal

Hinton’s proposed priority goes beyond measuring intelligence. He criticized big technology companies for continually wanting smarter systems, arguing that other characteristics deserve as much attention. His concern is not simply whether AI can perform more difficult tasks, but what kind of behavior accompanies that ability.

He described a gap between suppressing harmful behavior and developing a good character. In his view, training models on internet material gives them undesirable tendencies that later training teaches them to repress. He compared that process to producing a Victorian gentleman, rather than creating something humans could safely live alongside.

Hinton hopes that a country which learns to build genuinely benevolent AI would share that knowledge with allies and rivals. His rationale is that no country wants AI taking over elsewhere, creating a common interest in sharing safety knowledge even across political divisions.