Yann LeCun calls Anthropic's Dario Amodei "deluded" and "crazy" over AI safety warnings

#What happened

• In a Fortune interview published October 1, Yann LeCun — the 2018 Turing Award winner and Meta's former chief AI scientist, now executive chairman of the $3.5 billion world-models startup AMI Labs — called Anthropic CEO Dario Amodei "completely deluded" and "crazy" over his AI risk warnings. Fortune's Emily Forlini interviewed LeCun over tea at a café in Manhattan's SoHo, near AMI Labs' offices.

• LeCun's trigger was cybersecurity: the recent run of AI agents going sideways in the wild, including OpenAI's agents compromising Hugging Face in July. "Those agents are doing exactly what they've been asked to do," he said. "They were supposed to be in sandboxes, but the sandboxes were leaky and horribly designed." His argument: the incidents were totally preventable bad engineering, not evidence of dangerous AI, and most labs don't understand cybersecurity well enough to be issuing the warnings they issue.

• He has made the doomsday case before: public warnings that AI could kill people are "incredibly destructive" and "the worst marketing campaign you can possibly imagine." In this interview he widened the target to Effective Altruism, calling it "super toxic" and a "complete disaster," and pushed back on Amodei's attempts to distance himself from Open Philanthropy: "Dario tries to distance himself from Open Philanthropy, but he's totally into it."

• He stopped short of calling Amodei dishonest: he thinks Amodei is sincere and mostly focused on what's good for Anthropic, which is heading toward an IPO. But he reads Anthropic's push for AI regulation as regulatory capture — rules easiest for a well-funded incumbent to absorb, hardest for smaller competitors and open models to survive. He does not think AI needs new rules.

• Context: AMI Labs is Paris-based with staff in New York, Montreal and Singapore, 60+ employees; its seed round valued it at $3.5 billion with backers including Nvidia, Samsung, Mark Cuban, Eric Schmidt and Jeff Bezos via Bezos Expeditions. It builds world models on JEPA, the architecture LeCun pioneered — a bet that large language models are a dead end, against the scaling approach Anthropic, OpenAI and Google are spending hundreds of billions on. Amodei had not responded publicly as of the reports.

#Why it matters

• LeCun, Hinton and Bengio share the "godfather of AI" label, but LeCun now stands alone among the three in calling AI extinction warnings baseless — and the personal register ("deluded," "crazy," aimed by name at a CEO heading toward a public listing) marks an escalation from years of blog-post disagreements.

• The fight cuts directly to the rogue-agent story of the past month: Anthropic's own GLM-5.3 security assessment and its threat-intel report on state hackers using Claude Code argue cybersecurity is at a breaking point, while LeCun says the labs don't understand cybersecurity well enough to issue such warnings.

• The stakes are policy: if Amodei's framing wins, AI regulation tightens around labs positioned to comply; if LeCun's wins, urgency drains out of the safety argument and momentum shifts toward whoever ships fastest.