Meta Chief Executive Officer Mark Zuckerberg publicly criticized a closed ecosystem in which a small number of corporations such as OpenAI and Anthropic monopolize AI technology and control. At the same time, Meta returned to strengthening an open ecosystem by releasing open-source AI models that can run on personal devices.
Zuckerberg, in a 6,500-word essay released on the 10th, "The future belongs to everyone," said, "Rather than centralizing superintelligence, we should distribute it widely so everyone can control it," and argued, "The idea that AI is so dangerous that the only way to make it safe is to concentrate power to the extreme is fundamentally flawed."
Zuckerberg gave the example that a world where everyone can use a superintelligent lawyer would be fairer and more efficient than a world where only one person has a superintelligent lawyer. Similarly, he emphasized that if only one corporation holds superintelligence, market dynamism and prosperity could fall below current levels.
He added, "I don't understand why people who believe AI will eliminate most jobs and cause humanity to lose a significant part of its purpose are rushing to create such a future." The remarks are seen as aimed at OpenAI and Anthropic, which are emphasizing AI's potential risks while continuing development competition.
Zuckerberg also warned that a policy to delay the release of U.S.-made closed AI models by a month would pose significant risks to U.S. dominance in AI. Instead, he suggested that reviews of risks such as cybersecurity and the potential misuse for developing biochemical weapons should be carried out more proactively.
"My alternative is that, instead of the government reviewing a model after it is ready for release, we provide the government with intermediate training checkpoints of new cutting-edge models and technical staff so it can harden core systems and secure defenses against emerging risks," he said. "Through this approach, the government can build capacity without creating power imbalances and without restricting or delaying individuals' access to superintelligence."
He also expressed the view that the "distillation" method—long defined as copying another AI's responses and infringing on intellectual property—should be approached openly.
Meanwhile, Meta unveiled a new next-generation open AI model, "Muse Glimmer," in line with Zuckerberg's announcement.
Specialized for on-device AI, Muse Glimmer is a small model with 30 billion parameters. It is designed to continuously run AI agent tasks on a standard PC with a single graphics card.
Meta told The Wall Street Journal (WSJ) it also plans to introduce within weeks a model that releases the weights of its top-tier model, "Muse Spark 1.2."
Until last year, Meta had been developing the open AI model Llama but dropped the plan and returned to developing closed AI models after performance fell short of expectations. The renewed push for an open AI strategy is seen as both using the open ecosystem as a differentiator in a market led by OpenAI and Anthropic and responding to the spread of China's open AI models.
To dispel safety concerns about open models, Zuckerberg decided to establish a new governance structure that grants an internal independent board of directors the authority to review AI model safety standards and control whether to release them. He also decided to create a $1 billion ("The future belongs to everyone") fund to build AI infrastructure in the United States and support local communities.