Zuckerberg Pushes Open-Source AI in New Manifesto
Meta CEO Mark Zuckerberg published a sweeping manifesto on Monday advocating for a distributed, open-source approach to artificial intelligence, arguing that concentrating superintelligence in the hands of a few institutions would lead to “outcomes that are less favorable for everyone else.” The announcement coincided with the release of Muse Glimmer, a 30-billion-parameter open-weight AI model designed to run AI agents locally on consumer hardware.
In the approximately 6,500-word essay titled “The Future is for Everyone: The Path to a Positive AI Future,” published on Meta’s website, Zuckerberg proposed a philosophy built on “individual empowerment as the source of prosperity, invention as the primary purpose of superintelligence, and balance of power as the foundation of safety.”
A Direct Challenge to Rivals
Zuckerberg’s manifesto takes direct aim at the safety-first approach championed by rivals OpenAI and Anthropic, who have argued that frontier AI should be developed carefully and kept controlled. Without naming them explicitly, Zuckerberg criticized the prevailing discourse in the industry.
“It is surprising that the discourse from many developing AI is so filled with doom,” he wrote. “I do not understand why anyone who believes that AI will eliminate most jobs and much of humanity’s relevance would rush to build that future.”
He further argued that the notion of concentrating AI power as a safety measure is fundamentally flawed. “The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic,” Zuckerberg wrote. “Historically, hoping that an absolute power will benevolently provide for humanity if sufficiently enlightened has not led to safe or positive outcomes.”
Zuckerberg also made veiled references to competitors, stating: “Most other labs are focused on building AI for companies, governments, or other institutions, so if those labs lead, then the balance of power will favor larger institutions over individuals.”
Muse Glimmer: Open Weights for Local AI
Meta’s release of Muse Glimmer marks a significant step in the company’s open-source strategy. The 30-billion-parameter model is released under the permissive Apache 2.0 license, allowing developers to download, modify, and build on it freely. It is designed to run on a Mac or PC with a single consumer GPU, enabling “always-on” local agent workflows that can operate “anywhere, anytime, with or without an internet connection.”
According to TechCrunch, Glimmer is designed to run AI agents that can perform multi-step tasks—like calling tools, writing and debugging code, working with files and screenshots, and executing tasks over extended workflows—locally on consumer hardware. The model supports text and images and was trained across more than 100 languages.
Meta also announced it would provide access to Muse Spark 1.2, a more powerful AI model, though this remains closed-weight. The dual strategy allows Meta to benefit from community innovation while maintaining a competitive advantage through its more powerful proprietary models.
The Open vs. Closed AI Debate
Meta has long been a proponent of open-sourcing its AI models, particularly through its Llama family, but has struggled to keep pace with rivals including OpenAI, Anthropic, and Google, as AP News reported. The company formed Meta Superintelligence Labs last year to propel itself back into the high-stakes AI race.
Chinese companies including DeepSeek, Alibaba, and Moonshot have emerged as formidable low-cost alternatives in open-weight models, delivering performance that rivals top US systems. Reuters noted that Zuckerberg’s statement marks the latest show of support for open-weight AI, which is gaining traction as businesses grow wary of ballooning AI bills and worry about recent cybersecurity incidents involving models from Anthropic, OpenAI, and Meta.
Zuckerberg called for US policy changes on distillation and data use in training, arguing that American labs face additional restrictions that give foreign competitors an advantage. “Foreign labs currently hold several advantages here since American labs have to comply with many additional restrictions on training data,” he wrote. “US policy must reduce this additional friction if we want American open source models to lead over time.”
Governance and Community Investment
In a notable governance innovation, Meta announced it is implementing a structure that gives its independent board of directors the power to approve safety criteria for releasing models. Zuckerberg acknowledged that while Meta remains a founder-controlled company, he does not believe it is in anyone’s best interest for a single individual to be the sole decision-maker on how superintelligence is deployed.
Meta also announced a $1 billion “Future Is For Everyone Fund” to support communities hosting its data centers. This comes amid significant community opposition to AI infrastructure expansion. As Indian Express reported, data centers remain widely unpopular, with 42 states witnessing 142 protests on a single day last month in the first coordinated nationwide effort against rapid AI infrastructure expansion.
Critical Perspectives
Despite the sweeping philosophical arguments, critics have noted a gap between Zuckerberg’s advocacy for openness and Meta’s actual product decisions. Firethering pointed out that “fourteen pages is a long argument for openness from a company that kept its most competitive model closed on the same day it published them.”
The analysis noted that while Muse Glimmer is genuinely useful for developers, Meta’s most powerful model, Muse Spark, remains behind a paywall. “The philosophy stays consistent,” the analysis concluded. “The timing of what gets released and when tells a different story.”
NY Post reported that Zuckerberg’s push comes as leading AI rivals have been sounding a somewhat pessimistic note. OpenAI recently paused some “internal activities” involving its new model Astra over critical cybersecurity risk concerns, while Anthropic’s Claude Mythos model sparked fears of an “AI doomsday” with warnings of catastrophic hacks if released publicly.
What’s Next
Meta plans to spend up to $145 billion in capital and up to $600 billion by 2028 on AI infrastructure, according to reporting cited by Indian Express. The company is also reportedly working on a next-generation model internally called “Watermelon,” though it remains unclear whether it will be released openly or kept closed.
Zuckerberg’s vision of “personal superintelligence”—AI agents that work 24/7 on behalf of individuals to manage relationships, health, career, finances, and more—represents a significant strategic bet on consumer AI rather than enterprise AI. Whether this approach helps Meta close the gap with OpenAI and Anthropic, and whether policymakers respond to his calls for changes to distillation and training data policies, will be key questions in the months ahead.
As Zuckerberg wrote: “Putting power in people’s hands to pursue their own aspirations is how humanity has made the most progress. Novel ideas and major steps forward rarely originate from established institutions alone.”
For now, the release of Muse Glimmer provides developers with a powerful new open-weight tool, while the broader philosophical debate over the future of AI—open or closed, distributed or concentrated—continues to shape the industry’s trajectory.