In a New York Times interview with Ezra Klein published September 23, Nvidia CEO Jensen Huang dismissed the loudest AI extinction warnings as unscientific and said the predictions themselves are causing harm. He is far from the only prominent figure making some version of that case. But line up the others who agree the fear is overblown, and they don't actually agree with each other about why.
What Huang Actually Argued
Huang's position is narrower than a blanket dismissal of risk. He called the frequently cited estimate that AI carries roughly a 10% chance of destroying society not grounded in science, and said predictions framed that way are actively hurtful, discouraging young people from studying the field at all. In the same conversation, he directly rebuked Geoffrey Hinton, the Nobel-adjacent computer scientist known as a co-father of modern AI, over Hinton's own risk warnings.
Huang isn't arguing for zero oversight, though. He said a lab that concludes it cannot contain its own experiments, and that failure would cause real damage, should be shut down, comparing an unready AI system to a self-driving car that hasn't cleared safety testing. He separately told CBS there is a 0% chance AI ends the world by 2030 and said the industry should move as fast as it can. The common thread is evidence over forecasting: He wants safety decisions grounded in what a system has demonstrably done, not in speculative multi-year predictions.
LeCun: The Real Risk Is Concentration, Not Extinction
Yann LeCun, Turing Award winner and Meta's former chief AI scientist, has been publicly blunt with people predicting near-term AI catastrophe, and has compared panic over isolated bad incidents to blaming an entire industry for one bad actor's mistake. But LeCun's disagreement with the doom framing isn't that no real risk exists. His stated concern is that AI power becoming concentrated in a small number of companies or governments is the more serious danger, a different argument from Huang's evidence-over-forecasting position, even though both conclude the extinction framing is wrong.
Zuckerberg Broke From the Pacing Consensus Entirely
When Anthropic's Dario Amodei published his pacing essay this month, Sam Altman and Elon Musk both publicly agreed within days. Meta's Mark Zuckerberg didn't. He favored market-led safeguards over a coordinated industry slowdown, putting one of the four largest frontier labs outside the brief consensus that had formed around Amodei's proposal.
Andreessen: Move Fast, the Utopia Is Real
Venture capitalist Marc Andreessen has argued for years that AI will help save the world rather than end it, under a banner he calls effective accelerationism. His position predates this month's news cycle, but it remains institutionally significant: Andreessen has advised the Trump administration on AI policy, and fellow a16z partner Sriram Krishnan currently serves as the White House's senior AI adviser.
Gomez: The Extinction Story Is a Distraction, the Weapon Story Isn't
Cohere CEO Aidan Gomez is the most complicated case in this group, and worth reading carefully rather than sorting into either camp. He has called the idea of AI ending human existence an absurd distraction from real risks, and warned the public against buying into what he calls fearmongering.
Gomez's actual position is narrower than a blanket claim that the fear is exaggerated: The extinction narrative specifically is overblown and distracting, while cybersecurity misuse is a serious, present danger that industry self-regulation isn't equipped to handle alone. That's closer to Amodei's or Hinton's concern about concrete misuse than it is to Huang's or Andreessen's broader dismissal of doom framing generally.
Washington Has Its Own Reasons to Downplay Doom
Dean Ball, an AI policy fellow at George Mason University's Mercatus Center, offers a narrower institutional version of the same skepticism: Republican AI priorities in Washington right now, he argues, are data center buildout, government and military AI adoption, competing with China, and child safety, not existential risk. That's less a claim that the fear is scientifically wrong and more an observation that it isn't what's actually setting the legislative agenda.
Why the Distinction Matters
Treating AI doom skeptics as one side of a two-sided debate flattens what's actually a set of different claims with different implications. Huang's version implies safety regulation should wait for demonstrated failures. LeCun's implies the real regulatory target should be concentration of power, not capability itself. Gomez's implies aggressive investment in cyber-defense specifically, alongside independent oversight, while dismissing extinction talk as a distraction from that narrower fight. Andreessen's implies minimal intervention of any kind. These aren't interchangeable positions that happen to share a conclusion. Anyone weighing AI policy, or an AI vendor's own safety claims, against this debate should ask which specific argument is actually being made, not just which side of the doom line someone falls on.
Sources: AI Pulse · Big Picture · workplaceai.ai. Huang's New York Times interview, published September 23, 2026: coverage via Tech Insider, Seoul Economic Daily, TechRadar, and Tom's Hardware's reporting on a related CBS interview. LeCun and the broader "not everyone thinks AI will kill us all" roundup: CNN Business, September 24, 2026. Zuckerberg's position: BNN Bloomberg, September 21, 2026. Andreessen: TechCrunch. Gomez's positions: CNBC's "The Tech Download," CBC News, and the Financial Times. Dean Ball's analysis: TechCrunch, via BNN Bloomberg's broader roundup. Context on Linthicum's separate argument: previously reported by WorkplaceAI AI Pulse. Every quote and detail above is attributed to its original reporting; none is a WorkplaceAI study, and WorkplaceAI takes no position on which argument is correct.