Meta had long been the flagbearer of US’s open-source AI space, but it now seems to be rethinking its stance on open models.
The stark admission comes from Yann LeCun, Meta’s former Chief AI Scientist, who recently left the company to launch his own venture, AMI, focused on building non-LLM world models. Speaking candidly about the shifting landscape of AI development at the AI Pulse conference on 4th December, LeCun—a longtime advocate for open-source AI—painted a troubling picture of America’s retreat from openness, even as China doubles down on it. His comments suggest a fundamental shift in Meta’s approach, one that could reshape the global AI competitive landscape.
“I’ve been a big proponent for open source for a long time, as many of you know,” LeCun began. “And I think what we are observing at the moment is most big American players in the domain are kind of climbing up open. OpenAI, of course, stopped being open quite a long time ago. Anthropic never was. Google is kind of somewhere in between, but mostly climbing up too.”
Then came the revelation about his former employer: “And Meta is kind of rethinking its strategy about open source. There’s still gonna be quite a lot of open source coming out of Meta, but maybe not as systematic as it used to be.”
LeCun contrasted this American retreat with China’s aggressive embrace of openness: “And so conversely what you’re observing is that China, on the other hand, is going open source all the way. So the best open source models at the moment in the LLM world are all Chinese. And it’s not good.”
The implications of LeCun’s assessment are significant. Meta’s Llama series had been positioned as the Western answer to closed models from OpenAI and Anthropic, providing researchers, startups, and developers with powerful AI tools without the constraints of API-only access. The company’s apparent pivot comes amid broader internal upheaval. Following the underwhelming performance of Llama 4, Meta restructured its AI efforts entirely, bringing in Alexandr Wang to lead the charge and aggressively poaching researchers from top labs including OpenAI.
LeCun’s departure to build AMI reflects his belief that large language models cannot achieve artificial general intelligence—a conviction that may have put him at odds with Meta’s LLM-centric strategy. His warning about Chinese dominance in open-source AI models highlights a paradox: as American tech giants increasingly lock down their AI capabilities behind proprietary walls, they may be ceding leadership in the very technologies that democratize AI development. The question now is whether Meta’s recalibration represents a temporary strategic adjustment or signals a permanent shift away from the open-source philosophy that once defined its AI ambitions.