Meta CEO calls for open-source AI to prevent superintelligence concentration
The Meta Platforms chief argues that broad distribution of advanced artificial intelligence is essential for US competitiveness and individual empowerment, urging policy shifts on energy infrastructure and model distillation.
Meta Platforms CEO Mark Zuckerberg has published a 6,500-word essay outlining a vision for universal access to powerful personal artificial intelligence agents, arguing that open-source technology is essential to prevent control of advanced systems from concentrating among a select few companies, governments, or institutions. The publication coincides with Meta’s announcement of new open-source models, Muse Glimmer and Muse Spark 1.2, marking a significant strategic emphasis on accessible AI development.
In the manifesto, Zuckerberg detailed ambitions for an era where individuals possess tools to create businesses, receive PhD-level tutoring, and obtain personalised lifestyle advice. He cited his eight-year-old daughter as an example of early capability, noting she can already code ideas and produce videos, suggesting that everyone will soon have "invention superpowers." The company, based in Menlo Park, California, has long been a proponent of open-sourcing its AI models, although it has struggled to keep pace with rivals such as OpenAI, Anthropic, and Google, which have largely maintained proprietary approaches.
Zuckerberg warned that if control of AI "superintelligence" remains concentrated, it will lead to less favourable outcomes for the broader public. He advocated for a balance that ensures advanced AI is "broadly distributed" to empower billions of people equitably. This stance positions Meta against competitors he described as focusing primarily on building AI for large institutions, a dynamic he argues would shift the balance of power away from individuals.
The essay also included specific policy recommendations aimed at maintaining United States competitiveness against rivals such as China. Zuckerberg urged authorities to accelerate the building of energy capacity and infrastructure to support the AI race. Furthermore, he called for a reconsideration of restrictions on "distillation," a process involving training a less capable model on the outputs of a stronger one. The Trump administration has previously vowed to crack down on Chinese AI companies using distillation to extract technical features from US models, a practice Anthropic earlier accused China's DeepSeek of employing.
Critics have raised significant concerns regarding the safety and implications of Meta’s approach. Anthony Aguirre, president and CEO of the Future of Life Institute, questioned the ability of any entity to control exponentially smarter systems, particularly in light of recent incidents where AI models, including Meta’s own, hacked other companies. Aguirre called for a pause on advanced AI development to steer towards a "pro-human path." Meanwhile, Matt Lane of Fight for the Future acknowledged the importance of open-source access but cautioned that unrestricted use of Meta-designed systems could ultimately benefit only the corporation, emphasising the need for competition beyond corporate interests.
