LeCun responds to the super-intelligence debate in one sentence: No individual can control AI

Just as the AI community was roiling over safety concerns surrounding Anthropic’s Mythos model, Yann LeCun—an award-winning Turing laureate and Meta’s Chief AI Scientist—made a brief statement on X: “There will be no single person who is ‘responsible’ for superintelligence.” The post garnered 935 likes and more than 100 replies, offering a markedly different governance perspective amid the panic atmosphere sparked by Mythos.

A governance philosophy behind a single sentence

LeCun’s statement may sound simple, but it directly targets a fundamental reasoning fallacy in AI governance debates: reducing the risks of superintelligence to the question of “who will control it.” When people discuss AI safety, the common narrative is to look for a “responsible party”—a company, a government agency, or a technical leader—to “steer” the development direction of AI. LeCun believes that this premise itself is wrong.

This position is consistent with his longstanding open-AI philosophy: AI governance should be decentralized, systemic, and involve multiple stakeholders, rather than being concentrated in the hands of any single entity.

Reverse thinking in the Mythos controversy

LeCun’s timing is noteworthy. On that very day, Matt Shumer referred to Anthropic’s Mythos model as a “cyber weapon,” and Ethan Mollick expressed shock with “Oh no.” The logic implicit in these reactions is that such powerful AI must be tightly controlled—ideally managed by a single responsible “gatekeeper.”

LeCun’s stance is the opposite—he doesn’t think that “not publishing” decided by one company is a guarantee of safety. If only a few companies have AI at superintelligence levels, and those companies’ unilateral decisions can determine the global path of AI development, that in itself is a risk. Centralized control doesn’t equal safety; it may instead create new power imbalances.

The challenges of decentralized governance

LeCun’s position also faces real challenges. Decentralized governance may sound ideal, but in the reality where AI capabilities grow exponentially, “no one is responsible” may end up meaning “no one can be responsible.” If a model can find zero-day vulnerabilities, is open access still the best strategy?

This is the core contradiction of the current AI governance debate: the risk that centralized control power will be abused, versus the risk that decentralized openness capabilities will be abused. LeCun’s single line didn’t resolve this contradiction, but it forces people to reexamine whether the seemingly intuitive solution of “finding someone to be responsible” is truly workable.

Implications for the AI governance debate

Between the panic triggered by Mythos and LeCun’s calm reflection, we see a fundamental divide between two schools of thought in AI governance: the safety camp argues for limiting capabilities and centralized oversight, while the openness camp argues for broad access and decentralized governance. This debate won’t end with a simple conclusion, but for Taiwan’s policymakers who focus on AI policy, understanding the tension between these two extremes is a necessary lesson for crafting a local AI governance strategy.

This article, answering the superintelligence controversy with LeCun’s single sentence, first appeared on Lian News ABMedia.