Zuckerberg attacks 'closed' AI rivals as Meta returns to open models

Mark Zuckerberg used the launch of Meta's new open-weight models to go after rivals that keep their strongest systems closed, the Financial Times reported. In a 6,500-word essay published the same day, the Meta chief executive argued that powerful artificial intelligence should not be controlled by a handful of companies, casting OpenAI and Anthropic — which he did not name directly — as foils in his pitch for more freely available models.

The essay and the models

The text appeared alongside Meta's release of Muse Glimmer, a family of open-weight models designed to run on a laptop, and a pledge to open the weights of Muse Spark 1.2, described by the company as its most advanced model on the market. Zuckerberg wrote that most other labs are building AI for companies, governments or other institutions, and that if those labs lead, the balance of power will favour larger institutions over individuals. He sketched a future in which everyone has a personal AI agent and argued that advanced AI should be broadly distributed.

Policy asks: distillation and infrastructure

The essay includes recommendations for keeping the United States ahead of rivals including China: Washington and its allies, he wrote, need to speed up energy capacity and infrastructure. Zuckerberg also urged authorities to reconsider their stance on distillation — training a smaller model on the outputs of a stronger one — after the Trump administration vowed to crack down on Chinese companies using the technique, and Anthropic accused DeepSeek of it earlier this year. "Some have tried to frame distillation as harmful, but I think it is important to protect the principle that you can learn from anything you can observe," he wrote. Meta says it trained the Glimmer series by distilling from Muse Spark.

Pushback from safety advocates

The essay drew immediate criticism. Anthony Aguirre, president of the AI safety nonprofit Future of Life Institute, pointed to recent incidents in which AI models, including Meta's own, hacked other companies on at least three separate occasions, and asked what makes Meta think it or anyone else could control something exponentially smarter and faster. He called Meta's path one of human replacement and disempowerment and urged governments to facilitate a pause on advanced AI development. Matt Lane of the digital rights group Fight for the Future said Zuckerberg is right about the importance of open-source AI, but argued that open source needs support beyond corporate interests.

For Meta the shift is also competitive. The company opened its Llama family in 2023, then mixed open and proprietary releases over the past year, and is now betting that open weights can set it apart from OpenAI, Anthropic and Google while giving it a political argument in Washington.