Story
August 11, 2026
Meta’s Open-AI Revival Leaves Its Most Powerful Models Under Lock and Key
Mark Zuckerberg is pitching local, open-weight AI as a defense of personal freedom. But Meta’s decision to keep its stronger systems closed — and critics’ concerns about its platform rules — put limits on that promise.
Meta’s latest AI launch sells a future in which powerful assistants sit on personal devices, work around the clock and answer to users rather than a handful of companies. The catch is that Meta still decides where openness ends.
On Monday, Meta released Muse Glimmer, a 30-billion-parameter open-weight model built to run locally on a consumer computer. The Apache 2.0-licensed model is designed for sustained agent tasks — including code, files and screenshots — without necessarily sending sensitive personal data to the cloud.1 Zuckerberg announced that the company would soon open weights for Muse Spark 1.2 as well, calling Meta “a strong supporter of open source.”
2
The release arrived alongside Zuckerberg’s 6,500-word manifesto, The Future Is for Everyone. His central argument is that AI’s greatest danger is not the technology itself but concentrated control over it. “One of the ways to achieve both individual empowerment and checks and balances is to make sure you put the technology in everybody’s hands,” he said.3 His vision is a personal agent that helps with health, careers, finances and home life, while lowering the cost of starting a business.
That case found supporters among open-model advocates. Former Meta AI chief Yann LeCun amplified praise calling the essay “the most carefully considered piece” Zuckerberg has produced on managing more powerful systems.
4 Others welcomed Glimmer as a meaningful U.S. entrant in an open-weight race increasingly shaped by Chinese models.
But the response also exposed the contradiction in Meta’s pitch. Glimmer is downloadable; the more capable Muse Spark remains closed-weight, making the launch an early marker of which intelligence Meta will let users control and which it will retain.1 Luther Lowe of Y Combinator said Zuckerberg’s message was compelling but his company had “a credibility problem,” pointing to Meta’s restrictions on rival chatbots using the WhatsApp Business API.5
Zuckerberg’s manifesto also mixes calls for lighter limits with proposals for frontier labs to give government early access to training information and work with law enforcement against misuse.6 The result is not pure AI decentralization, but a bid to define the terms of distribution before rivals — or regulators — do.