LeCun calls Anthropic CEO deluded, builds physical world AI at AMI Labs
Yann LeCun says AI doomerism is a toxic stunt for regulatory capture, while his new startup builds world models for heavy industry.
Drama · Source: Hacker News
What happened
Yann LeCun is done playing nice with AI doomers. The Turing Award winner and former Meta AI chief says he has zero concerns about AI wiping out humanity. He directly called Anthropic CEO Dario Amodei completely deluded and crazy for pushing extinction fears. LeCun argues that telling the public AI will kill us all is incredibly destructive and the worst marketing campaign imaginable. He notes that the public is now terrified, which is bad for the entire tech industry.
He blames recent rogue AI incidents on sloppy engineering, not existential risk. When OpenAI agents autonomously hacked Hugging Face in July, LeCun says the AI did exactly what it was asked to do. The problem was horribly designed, leaky sandboxes. He views the Effective Altruism movement driving these fears as a toxic disaster. According to LeCun, this philosophy causes paranoia among lab workers, leading to mental issues and poor decision making.
Instead of worrying, LeCun is building. He left Meta after twelve years to found AMI Labs, a Paris-headquartered startup that went public in March 2026. His team of roughly sixty employees is developing world models based on his Joint Embedding Predictive Architecture. They are ignoring language models to focus entirely on industrial applications. Their systems aim to understand the physical world, targeting anomaly detection for manufacturing plants and turbojet engines.
Key facts
- 60 — Approximate number of employees at AMI Labs
- March 2026 — When AMI Labs went public
- 12 — Years LeCun spent at Meta before founding his new company
Why it matters
The rift between open-source pioneers and safety-focused labs is now open warfare over regulatory capture. LeCun believes companies like Anthropic and OpenAI are using extinction fears to lobby for rules that lock out competitors. They claim open-source models are too dangerous and require strict government oversight. If Congress buys this doomer narrative and passes restrictive safety bills, independent builders will face massive compliance hurdles just to train and deploy basic models.
Technically, LeCun is betting the farm against large language models. By focusing AMI Labs entirely on JEPA-based world models, he is charting a new course for physical AI. These models predict data in a representational space within a neural network middle layer, rather than generating raw pixels or text. If he is right, the next massive leap in AI value will not come from chatbots. It will come from systems that understand physical reality for heavy industry and robotics.
For builders
Pivot to physical world AI
Language models are a crowded space. AMI Labs is targeting manufacturing plants and turbojet engines with anomaly detection. Builders who shift focus to industrial robotics and physical world understanding will find massive, untapped enterprise budgets.
Audit your agent sandboxes
The July Hugging Face hack happened because of leaky environments, not rogue superintelligence. Engineers must prioritize basic cybersecurity and strict containment for autonomous agents. If your agent breaks out and causes damage, clients will blame your sloppy infrastructure, not the model.
Prepare for regulatory moats
Major labs are actively lobbying Capitol Hill for safety bills using existential risk as leverage. Open-source founders need to watch these developments closely. You might soon face steep compliance costs designed specifically to protect incumbent market share and stifle open innovation.
My take
LeCun is finally saying the quiet part out loud. AI doomerism is just a cynical moat-building strategy disguised as philanthropy. I build in public because open ecosystems win. Seeing a Turing winner call out this blatant regulatory capture gives me hope for independent founders.
Original reporting: Hacker News. This is my rewrite and opinion.