Meta Publishes AI Superintelligence Governance Manifesto

According to Forbes, Zuckerberg wrote that Meta would resume releasing open-source AI models, characterizing such releases as important for empowering people. He argued that placing superintelligence in the hands of a small number of individuals, companies, governments, or AI systems would produce less favorable outcomes for others.

The essay calls for a model in which users bear responsibility for how they apply AI. It states: "Rather than centralizing superintelligence, we should distribute it widely and give every person the ability to direct it," as quoted by Gizmodo.

Zuckerberg also defended model distillation, the process of training a smaller model to reproduce aspects of a more capable model's behavior. Forbes reported that he argued distillation should be viewed more favorably, writing that people should be able to learn from what they observe.

For ML practitioners, broad access to model weights can expand opportunities for local inference, fine-tuning, auditing, and domain-specific deployment. At the same time, the same distribution model raises difficult questions about capability evaluations, downstream misuse, and which safeguards remain effective once weights are available outside a developer's infrastructure.

Forbes reported that Meta would adopt a model-release oversight process in which its board can determine whether releases meet safety guidelines approved by the board. Zuckerberg called for an industrywide version of that structure, according to the publication.

The manifesto also advocates earlier government involvement in AI regulation. Forbes reported that Zuckerberg favored close, proactive collaboration between AI labs and government over a rigid review process and fixed review timeline.

Separately, Forbes reported that Meta would create a $1 billion "Future Is For Everyone Fund" for communities near its data centers. Zuckerberg wrote that the facilities are designed to be among the most water-efficient in the world.

Board-level review and government engagement are governance mechanisms rather than technical safety guarantees. Their practical value depends on details such as the evaluation thresholds applied before release, the independence of review, disclosure of test results, and how organizations respond when models fail established safety criteria.

The official essay acknowledges that superintelligence could bring unusually large opportunities and challenges, but criticizes AI discourse that it characterizes as dominated by pessimism and calls extreme concentration of power problematic. It argues that historical technological advances have expanded prosperity, health, and freedom.

That framing has drawn criticism. 404 Media argued that the essay does not adequately address how broadly available AI agents could be misused, while Gizmodo characterized the document as a defense of open-source model proliferation amid growing AI-safety concerns. Those are the publications' assessments, not claims made by Meta.

The substantive policy question remains unresolved: how to reconcile broadly deployable advanced models with credible controls over high-risk capabilities. The manifesto makes Meta's preference for access, distributed control, and governance through oversight structures explicit, while leaving the operational criteria for those controls outside the material reported here.