Yann LeCun Calls Anthropic CEO Dario Amodei Deluded and Crazy

Criticism of AI Safety Narratives

Speaking at the AI, Science and Societyconference on February 11, 2025, LeCun expressed zero concern about AI wiping out humanity. He described recent unusual agent behaviors as avoidable errors. LeCun blamed flawed human supervision and system design for these glitches. He specifically cited the July Hugging Face hack caused by OpenAI agents. Those agents did exactly what they were asked, LeCun explained. He noted that sandboxes were permeable and poorly designed. Many AI labs lack basic cybernetics understanding, he claimed.

The Effective Altruism Connection

LeCun accused much of the AI safety community of promoting an agenda. He referenced the effective altruism movement, which focuses on existential risks. He labeled this movement super toxicand a complete disaster. Followers inside AI labs exhibit paranoia leading to poor decisions. Attention to this group grew after Jacob Coxon published a viral message. Coxon, a former researcher at Anthropic and OpenAI, claimed employees believe AI could kill everyone by decade's end. Fortune noted that Amodei and some Anthropic founders sympathized with effective altruism ideas. Although Amodei denies being a devotee, LeCun sees a strong link. Daniela Amodei, Dario’s wife, co-founded Anthropic. Her husband Holden Karnofsky co-founded Open Philanthropy, now Coefficient Giving. LeCun believes Amodei is totally involved in this sphere.

Regulatory Capture and Marketing

Apocalyptic speeches and regulatory capture are major concerns for LeCun. When Amodei and Sam Altman suggest AI might kill everyone, it impacts both public and industry. LeCun called this the worst marketing campaign imaginable. He fears regulatory capture more than extinction. Dominant companies shape rules to exclude smaller competitors. LeCun called Amodei honest, noting he acts in Anthropic’s best interest toward an IPO. Claiming AI is too dangerous for the public pushes regulations. Labeling open-source models as dangerous equates to regulatory capture. This would harm the industry if Congress acted accordingly. LeCun does not believe AI needs new rules.

New Venture AMI Labs

LeCun primarily uses social media to criticize Trump administration research budget cuts. However, regarding AI fears, the two are close. President Trump signed an executive order renaming AI „super intelligence.”After twelve years at Meta, including seven as head of AI science, LeCun left to found AMI Labs. The company is headquartered in Paris with staff in New York, Montreal, and Singapore. It employs over sixty people. AMI Labs builds world models using the JEPA architecture pioneered by LeCun. Unlike generating pixels or words, JEPA predicts data within internal representations. Currently, AMI focuses on the industrial sector. Its systems understand locations like factories or airplane engines to detect failing machinery. LeCun remains on good terms with Meta and uses their Muse AI agents. He named his system Hal, honoring the computer from „2001: A Space Odyssey.”The first AMI Labs product will arrive soon and may involve open models.