Jensen Huang has never lacked confidence. This week he reached for the most absolute figure in probability.
Asked about warnings that AI could end humanity by the end of the decade, the Nvidia chief executive told CBS News that “2030 is not going to be the end of the world. There is 0% chance that’s going to be the end of the world.” He added that scaring people is “unnecessary” and “irresponsible.”
A zero is not a forecast. It is a refusal to do the math. That makes it an odd centerpiece for the man who sells the machinery behind the most consequential technology race in history.
What set him off
Huang was responding to former Anthropic researcher Jacob Coxon, who claimed on social media that AI developers “earnestly believe that it could kill us all by the end of the decade.” Coxon said he had worked on pre-training research at OpenAI from 2023 to mid-2026 before a short stint at Anthropic, and he argued on his way out that neither company acts responsibly. Two other researchers reportedly backed him.
The timing mattered. Coxon’s posts drew hundreds of thousands of engagements and pushed AI oversight into mainstream debate.
The fight was already underway
Huang did not walk into a calm room. On September 12, Anthropic CEO Dario Amodei published an essay calling for a slowdown in American and global AI development, arguing that regulation targeting all US frontier labs works best because it also covers companies unwilling to cooperate voluntarily. Sam Altman and Elon Musk backed him.
Anthropic had already moved in this direction internally through its Responsible Scaling Policy and ASL-3 framework, which ties security requirements to capability thresholds rather than hard stops.
Then the two men shared a stage. At Dreamforce on September 15, Amodei pressed for globally agreed standards while Huang countered that market forces handle it and no new rules are needed. Huang called slowing down a “false choice.”
The real argument isn’t about extinction
Read the interview closely. Huang spends less time on doomsday and more on jurisdiction.
He rejected new regulations outright, saying existing US product liability and unauthorized-entry laws already cover the ground, and said he agrees with President Trump that the industry needs no additional guardrails. His warning was blunt: apply current law first, and don’t let a doomsday narrative excuse anyone from the rules already on the books.
That argument deserves a hearing. It is also a different argument.
Whether a technology can end the world is a scientific question. Whether liability law written for defective products can govern a system that acts autonomously, at scale, across borders, is a legal one. The second question remains wide open. The Center for AI Safety and academic researchers have spent years on exactly that gap.
Why the messenger matters
Huang’s position did not form in a vacuum. During the All-In Summit on September 14, Trump phoned him on stage, told him “the robots will not be taking over,” and described AI fears as a “hoax.”
Meanwhile, Nvidia now carries a $5 trillion valuation and earns from every model trained, cautious or reckless. Nobody needs to accuse Huang of bad faith to see the structure. The company with the most to gain from acceleration is the one certifying acceleration as safe.
The reverse holds too. Labs calling for a slowdown compete for regulatory position, and Huang has said so before. Last year he dismissed Amodei’s job-loss forecasts with a sharper line: “He thinks AI is so scary, but only they should do it.” Critics like Neil Chilson of the Abundance Institute make the same point — safety rhetoric doubles as a pitch to regulators that only the largest labs can handle the work.
Amodei rejects that reading. He has called the accusation an “outrageous lie” and bristles at being labeled a doomer.
Neither camp is a neutral witness. Both have money riding on the outcome.
The takeaway
Frame this as Huang versus the doomers, and you miss it.
The AI safety debate has stopped being technical. It is now a fight over who carries the burden of proof. Safety researchers say builders must show a system is controllable before scaling it. Huang says critics must show harm before anyone constrains it.
That split runs through the whole field. Expert disagreement on existential risk has less to do with new evidence than with which default applies when evidence runs out — a divide visible again in Altman’s own warnings about loss of control and concentrated power.
Nobody has proposed a way to settle it. Until someone does, confident numbers will keep substituting for evidence that doesn’t exist yet. “>10%” and “0%” fail the same test.
The honest answer to “how likely is catastrophe?” is that we don’t know. Building policy on that uncertainty is harder than picking a number — which is precisely why so few people in this debate bother.
Related: Anthropic AI Development Metrics: Claude Leads 26% of R&D
