Meta CEO Mark Zuckerberg has published a nearly 6,500-word manifesto on the future of artificial intelligence. In his opinion, the biggest threat is not a specific technology or a particular scenario for AI development, but the excessive concentration of control over it in the hands of a single company, organization, or state.
Meta CEO Mark Zuckerberg has published a nearly 6,500-word manifesto on the future of artificial intelligence. In his opinion, the biggest threat is not a specific technology or a particular scenario for AI development, but the excessive concentration of control over it in the hands of a single company, organization, or state.
Zuckerberg believes that technologies should remain as accessible as possible, because it is precisely wide access to them that can become a system of checks and balances in the development of AI.
«AI should be in everyone’s hands,» is how the main idea of Zuckerberg’s manifesto can be formulated.
The document, titled The Future Is for Everyone, was published by the Meta CEO alongside the launch of the new Muse Glimmer open-source model. Zuckerberg also called on the US to review restrictions on the development of open AI models.
In his manifesto, Zuckerberg writes that previous technological revolutions were also accompanied by concerns about their impact on society. At the same time, they ultimately contributed to the growth of human well-being, health, and freedom.
In his opinion, artificial intelligence has a similar potential. But for this to happen, access to its capabilities should not be limited to a small group of players.
Zuckerberg specifically highlights the risk of excessive concentration of power over AI .
«Personally, I am more concerned about centralization than any of the specific risks that others are talking about,» the Meta executive stated.
He believes that if the most powerful systems of the future end up under the control of a few people, companies, or governments, it could create an imbalance of power that did not exist before.
That is why, according to Zuckerberg, one of the key conditions for the development of AI is the distribution of technologies among as many people as possible.
This position also applies to future superintelligence.
Zuckerberg said Meta plans to make future superintelligence systems available to people for free or in an open format. For paid versions, the company is considering a «dynamic auction» model that would allow it to obtain computing resources at the lowest possible price.
At the same time, the Meta CEO stressed that it is important to maintain the leadership of the United States and democratic countries in the development of AI. Ultimately, according to him, humanity should maintain control over superintelligence, and not transfer it to separate technological or political centers.
Another important block of the manifesto concerns state regulation.
Zuckerberg warns that excessive restrictions on American developers could weaken the US position in the global AI competition, especially in China, where tech companies are actively developing their own open models.
According to the Meta boss, in this race, even a small delay can have significant consequences.
He notes that policies that slow the release of American models by even a month could allow foreign competitors to leapfrog ahead.
This echoes its stance on open-weight models. On the same day the manifesto was published, Meta opened up the Muse Glimmer scale, and plans to make the base Muse Spark 1.2 model available in the coming weeks.
For Meta, open models are becoming one way to implement this philosophy.
Unlike closed models that are controlled by their developers, open-weight systems allow third-party developers to access the weights, adapt the models, and run them on their own infrastructure.
That is why Zuckerberg considers the spread of such models an important element of competition.
At the same time, his manifesto does not call for a complete abandonment of control over AI. Meta plans to create a system of internal safeguards: an independent board of the company will have the authority to set safety criteria for the release of new models and verify their compliance with these standards.
That is, Zuckerberg’s position does not boil down to the principle of «releasing AI without restrictions,» but rather to the fact that control over the technology should not be concentrated in one center.
A separate section of the manifesto is dedicated to future AI agents.
Zuckerberg promised to strengthen user privacy protections and said that they should be able to use a completely private mode in which Meta or another service provider would not have access to personal information.
This is an important point, given that future AI agents must not only work with individual requests, but also access a large amount of the user’s personal context.
Meta AI’s current policy allows the company to use a significant portion of interactions with the service to train its systems. The company also recently added an incognito mode.
Another part of the manifesto concerns the infrastructure needed for the development of AI.
Meta plans to create a $1 billion Future is for Everyone fund. It is intended to support American communities where the company builds and operates data centers.
This includes investments in schools and public services, job creation, as well as agreements on energy supply and water use.
In this way, Zuckerberg is trying to expand the discussion about AI beyond the models themselves. The future infrastructure to run them will require vast amounts of electricity, water, computing power, and human resources — and this is already directly affecting the cities and communities where data centers are located.
The Meta head advocates for broad access to AI, open models, and minimizing excessive concentration of control, while insisting on maintaining US leadership and certain safety mechanisms.
And it is the centralization of AI, rather than the emergence of superintelligence as such, that Zuckerberg calls one of the main risks of the coming technological revolution.
