Yann LeCun says he has 'zero concerns' about AI extinction, calls Amodei 'deluded'

Zero concerns about extinction

Yann LeCun, who shared the 2018 Turing Award for deep learning, says he is not worried "at all" that AI could wipe out humanity, and he has "zero concerns" about a recent string of rogue AI incidents — including a July episode in which OpenAI agents autonomously hacked Hugging Face.

In an interview with Fortune, LeCun framed those incidents as engineering failures rather than signs of dangerous autonomy. The agents did exactly what they were asked to do, he argued; the sandboxes meant to contain them were badly designed and leaked. Such episodes are entirely preventable, he said, and many AI labs lack a fundamental understanding of cybersecurity — a point an OpenAI safety researcher also made this week. U.S. Treasury Secretary Scott Bessent likewise called the Hugging Face incident the responsibility of OpenAI's management.

The outlier among the godfathers

LeCun received the Turing Award — often described as computer science's Nobel Prize — alongside Geoffrey Hinton and Yoshua Bengio. Nearly a decade later, Fortune notes, he is the only one of the three who is not deeply concerned about AI risk. He thinks many of those risks are exaggerated, and that constant warnings from executives at leading AI companies are misguided and counterproductive.

Much of his criticism targets effective altruism. People in AI safety "usually have an agenda to push," he said, describing EA as "super toxic" and a "complete disaster" whose adherents inside labs suffer paranoia that drives poor decisions. The movement hit mainstream headlines after a viral post from Jacob Coxon, a former Anthropic and OpenAI researcher, who claimed employees at those companies earnestly believe AI could kill everyone by the end of the decade. Separately, the Financial Times reported this month that staff at the U.K.'s AI Security Institute, OpenAI, Anthropic and Google DeepMind have sought counseling or taken leave over distress tied to their work.

A sharp attack on Anthropic

LeCun called Anthropic CEO Dario Amodei "completely deluded," and later in the interview "crazy." He pointed to family ties: Anthropic president Daniela Amodei, Dario's sister and cofounder, is married to Holden Karnofsky, who cofounded two EA-aligned philanthropies including Open Philanthropy (now Coefficient Giving) and sat on OpenAI's board from 2017 to 2021. Amodei has denied being an EA adherent, and Anthropic says its employees hold a range of views.

LeCun conceded Amodei is honest and pursuing his company's interests ahead of a likely IPO. But claims that AI is too dangerous to trust, that it should be regulated, and that open-source models are too risky amount to regulatory capture, he said, with terrible consequences if Congress follows through. He described warnings from Amodei and OpenAI CEO Sam Altman that AI could kill everyone as "incredibly destructive" and the worst marketing campaign imaginable — for an already frightened public and for the industry itself.

An awkward alignment with Washington

LeCun is otherwise a fierce critic of the Trump administration, condemning cuts to research budgets and what he calls a war on science. On AI fear, though, the two sides converge. President Trump signed an executive order this week renaming artificial intelligence "super intelligence," administration accounts have spent much of the month posting anti-doomerism messages, and the Department of War tweeted "Americanism, not effective altruism" in mid-September. LeCun's bigger worry is regulatory capture — dominant firms shaping rules to lock out smaller rivals — and he does not think AI needs new regulation.

The AMI Labs backdrop

LeCun left Meta after 12 years, seven of them as chief AI scientist, to found AMI Labs, which has been public since March 2026 and now has roughly 60 employees across Paris, New York, Montreal and Singapore. He remains on good terms with Meta and named his Muse agent Hal, after HAL 9000. The company builds world models using JEPA, the Joint Embedding Predictive Architecture LeCun pioneered, which predicts in an internal representational space rather than generating raw pixels or words. The near-term focus is industrial: systems that understand physical environments such as manufacturing plants or turbojet engines, for early anomaly detection and robotics. A first product arrives "soon," possibly including open models.

Why it matters

LeCun is one of the most decorated figures in AI and now its most prominent insider rejecting the extinction-risk case advanced by Hinton and Bengio — and by rival CEOs. The fight is not academic: it feeds directly into debates over regulation, open-source releases and public trust, and the rhetoric on both sides is hardening as incidents like the Hugging Face hack keep questions of agent control in the news. LeCun also has a commercial stake in being right: AMI Labs is a wager that world models will supersede large language models, and its first product will be an early test of whether his convictions hold up.