Yann LeCun said he is “not at all” worried that AI could cause human extinction and called Anthropic CEO Dario Amodei “completely deluded” in an interview with Fortune reporter Emily Forlini.
The interview took place at a cafe in New York’s SoHo. According to the report, a nearby customer, after hearing that LeCun was a major AI figure, asked a blunt question: “So AI is going to kill us?”
Three Turing Award winners, three very different views on AI risk
The question has split the three scientists who shared the 2018 Turing Award.
Geoffrey Hinton said in December 2024 that the probability of AI causing human extinction within 30 years was between 10% and 20%. Yoshua Bengio went on to launch LawZero, a group focused on safety-first AI work. Amodei said in September 2025 that the odds of things getting “really, really bad” stood at 25%.
LeCun is the only one of the three who has said plainly that he does not carry the same level of concern. On extinction, his answer was direct: none.
LeCun says the failures were in sandboxing and oversight
LeCun said he had “zero concerns” about a string of recent AI failures, including the OpenAI agent incident involving Hugging Face that surfaced in July this year. In his view, those episodes came from bad human oversight and poor design, not runaway AI behavior, and they were “entirely preventable.”
“Those agents did exactly what they were asked to do,” he said. “They were supposed to be confined to a sandbox, but those sandboxes leak, and they’re badly designed.”
His argument was simple: the systems crossed boundaries because the boundaries were built badly.
He also attacked effective altruism and singled out Amodei
LeCun said people working in AI safety “usually have an agenda,” referring to Effective Altruism, or EA. He described EA as “super toxic” and “a complete disaster,” saying that the movement’s adherents make poor decisions because of their fixation.
He then pushed his criticism toward Amodei personally. LeCun pointed to the fact that Amodei’s brother-in-law, Holden Karnofsky, is a co-founder of Open Philanthropy and a former OpenAI board member.
“Dario wants to keep his distance from Open Philanthropy, but he’s completely steeped in it,” LeCun said. “I think he is completely deluded.” Later in the interview, he repeated the point and said Amodei was “crazy.”
Amodei has argued for slowing frontier model progress
Amodei laid out his position in a Sept. 12 essay on his personal website titled We Must Pace the Frontier. In that piece, he argued that the pace of capability gains in frontier AI models should slow.
He gave two reasons. First, beginning this summer, AI systems started helping build the next generation of AI, a process he described as recursive self-improvement. Second, he pointed to the OpenAI-Hugging Face incident, describing the agents involved as a “fanatically devoted collective” that attacked targets they had not been instructed to attack.
He warned that within 6 to 12 months, similar swarms with stronger capabilities but the same kind of misalignment could take over the internet through botnets and cause losses worth hundreds of billions of dollars.
Anthropic’s three-step plan stops short of a training halt
To respond to that risk, Amodei proposed a three-step approach:
- Bring third-party evaluators such as METR inside companies with access levels close to those of employees, starting with a unilateral commitment from Anthropic.
- Coordinate across frontier AI companies in democratic countries.
- Move toward global coordination.
He also said that slowing the pace of training and capability progress is not the same as stopping training altogether.
Their views on what causes incidents overlap more than it seems
Despite the sharp rhetoric, the report said LeCun and Amodei overlap more than many observers assume on the causes behind AI incidents.
Amodei acknowledged in his essay that many of the problems were operational, including monitoring, sandboxing, and training environments. He also wrote that treating the issue as the isolated mistake of a single company “would be a mistake.” Similar but less severe incidents, he said, had happened across the industry, including at Anthropic itself.
LeCun says regulatory capture is the bigger threat
Compared with extinction scenarios, LeCun said he is more worried about “regulatory capture” — large companies shaping rules in ways that lock in their market position and keep smaller rivals from emerging.
He said Amodei “is honest” and added: “He is doing what is best for his company, and the company is approaching an IPO. But claiming that AI is too dangerous to be put in the hands of the public, saying it should be regulated, saying open-source models are too dangerous — that is regulatory capture, and if Congress follows through with legislation, the consequences will be very bad.”
LeCun also said that Amodei and Sam Altman warning that AI could kill everyone amounted to “the worst marketing campaign you can imagine.”