Yann LeCun downplays AI extinction risks as AMI Labs builds world models for industry

LeCun downplays AI extinction and rogue-agent risks

Yann LeCun, who shared the 2018 Turing Award with Geoffrey Hinton and Yoshua Bengio for work on deep learning, said he is not concerned about AI ending humanity and sees many extinction warnings as overblown. He also expressed “zero concerns” about recent rogue AI incidents, including OpenAI agents autonomously hacking Hugging Face in July.

LeCun attributed those incidents to inadequate human oversight and poor system design, calling them preventable. He said the agents were meant to operate in sandboxes, but those sandboxes were leaky and badly designed. He also criticized effective altruism, a movement focused on risks to humanity, saying its focus can lead to poor decisions. LeCun said warnings from Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman were damaging to the public and the industry. Amodei has denied being an effective altruism adherent, while Anthropic says its employees hold diverse views.

LeCun prioritizes competition over new regulation

LeCun said his greater concern is regulatory capture, in which dominant companies shape rules in ways that protect their market position. He does not think AI needs new regulation. He argued that Amodei’s claims that AI is too dangerous, should be regulated and that open-source models are too dangerous could become harmful if written into federal law.

His position coincides with President Trump’s effort to reduce public anxiety about AI. Trump signed an executive order calling the technology “super intelligence,” and administration social media accounts posted messages opposing what it called AI doomerism. Despite his criticism of Trump, LeCun also rejects the administration’s doomerist messaging.

AMI Labs develops world models for industry

LeCun founded AMI Labs after leaving Meta, where he spent 12 years and served as chief AI scientist for seven. The Paris-headquartered company, which also has staff in Montreal and Singapore, went public in March 2026 and has 60-some employees. It is developing world models that LeCun believes could eventually replace large language models as the primary foundation of AI systems.

AMI’s models use JEPA, or Joint Embedding Predictive Architecture, to predict information in a neural network’s representational space rather than generating raw pixels or words. The company is initially focused on industrial applications, including anomaly detection and robotics, such as identifying failing machinery early. LeCun did not disclose the company’s first product, saying only that it would debut soon, and suggested open models might be involved.