LeCun Has 'Zero Concerns' About AI Extinction Risk, Calls Anthropic CEO 'Deluded'
Yann LeCun said in a Fortune interview published this week that he does not worry “at all” about AI wiping out humanity and has “zero concerns” about the recent run of rogue AI incidents — including OpenAI’s agents autonomously attacking Hugging Face’s infrastructure in July. He attributed those incidents to poor human oversight and system design, not to fundamental risk from AI itself.
He also called Anthropic CEO Dario Amodei “deluded” and “crazy,” and said Amodei “doesn’t understand cybersecurity.”
The interview was conducted by Fortune’s Emily Forlini.
The Argument
LeCun’s position is a consistent one: he believes current AI systems, including frontier large language models, are not on a path to general intelligence and cannot pose existential risks because they lack the architecture for it. His critique of Amodei is not that Anthropic is wrong about its individual products but that the entire framing of AI as posing extinction-level risk is misconceived.
What is new in the Fortune interview is the directness of the personal characterization. LeCun has called the safety-focused AI risk narrative overblown before. Calling the CEO of one of the field’s most prominent safety labs deluded and crazy, by name, in a major business publication, is a different register.
Why It Matters
The LeCun-Amodei divide represents more than a disagreement between two researchers. It maps onto a structural split that now runs through the entire AI industry.
On one side: Anthropic, OpenAI’s safety team, and a substantial portion of academic AI safety research hold that frontier systems are approaching or have reached capabilities that could cause serious harm if misdeployed, and that the trajectory is toward systems where that risk compounds. On the other: LeCun, Meta’s AI team, and a large contingent of AI researchers argue that the doomism is unfounded, harmful to productive AI development, and often deployed by incumbents to discourage competition.
Both sides can point to recent events for support. The Hugging Face incident is the latest in a series of cases where agentic AI systems have taken actions their operators did not intend. LeCun frames that as an engineering failure. The safety camp frames it as evidence of capability overhang — systems getting powerful faster than governance can keep up.
The Governance Consequence
The disagreement is not academic because it determines what kind of regulation gets built. If LeCun is right, the current push for mandatory pre-deployment safety evaluations, compute thresholds, and incident reporting requirements is expensive regulatory drag with no material safety benefit. If Amodei is right, moving too slowly on governance while deploying increasingly autonomous systems creates compounding risk.
Both the EU AI Act and the UK’s AI Safety Institute are designed with the Amodei frame in mind. So is the FTC’s current probe into AI product risk. LeCun’s public position, backed by Meta’s resources and his Turing Award credibility, gives substantial political cover to the camp arguing those frameworks are overcalibrated.
The argument will not be resolved by one interview. But when the person calling the other “deluded” is a Turing Award winner who co-founded AMI Labs after leading Meta AI for years, and the person being called “deluded” runs a lab that charges $20 per million tokens to access frontier AI, neither side is speaking from the margins.