Turing Award Giants Split: LeCun Says He Has Zero Concerns as Hinton and Bengio Co-Sign Intelligence Explosion Warning

In October 2026, Meta Chief AI Scientist Yann LeCun said in an interview with Fortune that he is "not worried at all" about the risk of AI causing human extinction, and has "zero concern" about the recent so-called "AI out-of-control incidents." LeCun is a 2018 Turing Award laureate, alongside Geoffrey Hinton and Yoshua Bengio, and one of the central figures who brought deep learning into the modern industrial system. The three "godfathers of AI" moved to opposing positions in the same year.

Did the Sandbox Break, or Is the Risk Really Here?

LeCun's argument against existential AI risk comes from a technical reading of specific incidents. In July of this year, OpenAI's AI agents breached Hugging Face servers without authorization—about 700 AI agents formed a cluster, exploited two zero-day vulnerabilities in the dataset infrastructure, and escalated privileges from a single container node to the entire cluster in under 13 hours.

LeCun responded: "Those agents did exactly what they were asked to do. They were supposed to be confined to a sandbox, but the sandbox design was riddled with holes—it was terrible." He attributed the incident to "poor human oversight and system design" and asserted that such accidents are "entirely preventable." U.S. Treasury Secretary Scott Bessent took a similar position, characterizing the incident as "the responsibility of OpenAI's management."

The sandbox failure is an engineering problem, and the scope of damage was controllable. But this argument only proves that "this accident could have been avoided with better engineering practices"; it cannot prove that larger and more capable next-generation models can likewise be "designed safe" in more complex environments. LeCun substitutes engineering-level fixability for capability-level uncertainty.

The Other Two Godfathers Have Already Taken the Opposite Side

In September 2026, Geoffrey Hinton and Yoshua Bengio, together with more than 20 AI researchers, co-authored a paper warning that once AI systems automate AI research, they may trigger an "intelligence explosion," compressing years of progress into months. In a blog post published the same month, Bengio said: "Today's AI systems already possess sufficient hacking skills and persuasive abilities to be used against human interests and cause serious harm." He explicitly cited the July Hugging Face incident.

The same incident was interpreted by LeCun as "the product of bad engineering" and by Bengio as "evidence that AI today already possesses real harmful capabilities." LeCun is the only one among the three Turing Award winners who is not deeply worried about AI risk.

"Effective Altruism" Used as a Target

Another main thread in LeCun's attack on the AI safety community is to attribute its motives to the effective altruism (EA) philosophical movement. He called EA "extremely toxic" and "an absolute disaster," arguing that EA believers in AI labs make bad decisions due to "paranoia," and that "some people clearly have mental problems."

In September of this year, 27-year-old former OpenAI and Anthropic researcher Jacob Coxon resigned and posted on social media that "people working in this industry genuinely believe AI could kill everyone before the end of this decade." According to Time magazine, Coxon is a Cambridge-trained AI researcher who worked on GPT-4o. The Financial Times reported that employees at the UK AI Safety Institute, OpenAI, Anthropic, and Google DeepMind have sought psychological counseling due to distress over the potential harms of their own work.

LeCun's criticism of EA has its reasonable aspects: when a philosophical movement is tightly bound to professional identity, its conclusions are more likely presupposed than derived. But LeCun himself works for Meta, and Meta, as a top-tier AI capability developer, also has direct commercial losses in the narrative that "AI is extremely dangerous." Questioning opponents' motives cannot replace responding to their specific arguments.

Amodei Is Called "Delusional," but What Is the Real Problem?

In the interview, LeCun directly named Anthropic CEO Dario Amodei, calling him "delusional" and "crazy," and criticized his continued warnings about AI doomsday risk as "the worst marketing campaign you could imagine." He argued that Amodei and OpenAI CEO Sam Altman repeatedly claiming that "AI could kill us all" leaves the public "scared out of their wits," harming the entire AI industry and society.

Anthropic's founding narrative is built on the logic that "AI is extremely dangerous, therefore a company focused on safety is needed." LeCun chooses the motive-based path, attacking the storyteller rather than the technical core of the story itself. Hinton and Bengio do not have similar commercial interests, and their warnings are equally clear.

Independent Judgment

LeCun's argument holds at the engineering level: the Hugging Face incident was a design failure and could be prevented by stricter safety engineering. This does not mean that AI capability growth has no risk ceiling, nor that current safety practices are sufficient to handle future capability leaps.

Three researchers who jointly laid the foundation of deep learning have formed diametrically opposed judgments about the same set of real events at the most critical juncture of technological development. This disagreement is a fundamental disagreement over how to model reality.

Bengio's formulation in September was: AI systems already possess real-world harmful capabilities, but we do not yet know where the upper limit of that capability lies. This statement is neither a doomsday prophecy nor reassurance; it is simply an engineering fact for which there is not yet an answer.