Mark Zuckerberg has stepped into one of Silicon Valley’s most consequential debates over artificial intelligence, warning that concentrating control of advanced AI systems among a small number of companies could restrict innovation and place too much technological power in too few hands.
In an interview Tuesday with The New York Times, the 42-year-old Meta chief executive did not explicitly name Anthropic or OpenAI, two of the world’s most valuable AI startups. But his criticism was clearly aimed at the approach taken by companies that argue the most capable AI systems are too potentially dangerous to be released or developed without strict safeguards and centralized control.
Zuckerberg argued that tightly controlling the development of advanced AI would represent a departure from the broader tradition of American technology development, in which increasingly powerful tools have often become accessible to large numbers of people and businesses.
“So much of the discourse from a lot of the other labs that are developing this is overwhelmingly filled with doom,” Zuckerberg said. “There needs to be a voice or several voices that are bringing realism to this debate.”
His remarks deepen an increasingly public ideological divide within the technology industry over who should control the next generation of AI and how much access developers, businesses and individuals should have to increasingly powerful systems.
At one end of the debate are executives such as Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, who have argued that advanced AI models could create serious risks if their capabilities are distributed without adequate safeguards. Their concerns include what researchers call alignment risk: the possibility that an AI system’s objectives or behavior could diverge from human interests.
On the other side are companies including Meta, Microsoft, Nvidia and Google, which have supported broader access to AI technologies and, in varying degrees, open-source development. Their argument is that wider access allows researchers and developers to examine, improve and build upon AI systems, potentially accelerating innovation and creating new businesses.
The disagreement has taken on greater urgency as Chinese AI advances have challenged the technological lead that American companies once appeared to enjoy. The emergence of powerful Chinese models has also intensified debate in Washington over whether unrestricted access to open-source AI could create national security risks.
American technology executives have increasingly taken that argument to policymakers. Nvidia CEO Jensen Huang recently published a letter supporting open-source software development, attracting support from other technology leaders. He subsequently announced a coalition focused on open-source safety and cybersecurity tools for AI, with companies including SpaceX and IBM joining the effort.
Executives including Huang and Altman have been in Washington this week as the Trump administration considers how to regulate AI and address the technology’s national security implications.
Zuckerberg has positioned Meta increasingly clearly on the side of broader access. His argument is not simply that more people should be allowed to use AI but that the technology should ultimately be personalized rather than controlled by a single centralized system that claims to represent humanity’s interests.
“I think a lot of people have this notion that if you build some kind of singular AI you can, through some idealized form of alignment, make sure it is benevolent to humanity,” Zuckerberg said. “I’m personally skeptical of that path.”
Instead, he has promoted what he calls “personalized superintelligence,” in which AI systems would be tailored to individual users and their particular goals, preferences and values.
Zuckerberg said the idea of a single benevolent superintelligence is fundamentally difficult because people do not share identical values or political, cultural and personal priorities.
“I think it is literally impossible to have a single benevolent superintelligence that is simultaneously aligned with everyone at once,” he said.
That argument puts Zuckerberg at odds with a central concern among some AI safety researchers: that sufficiently powerful AI systems could eventually become difficult to control, making careful development and restricted access necessary to prevent catastrophic outcomes.
Amodei has been particularly vocal about AI safety and alignment risks. In a blog post Monday, he said Anthropic was focused on preventing powerful computing chips from reaching authoritarian governments, stopping industrial-scale distillation of AI models and requiring safety testing for sufficiently capable systems, whether they are open or closed.
The tension between the two philosophies was also evident Tuesday when senior AI researchers from Anthropic, OpenAI, Meta and Google signed an open letter titled “Pacing the Frontier.” The letter called for international efforts to develop technical and governance mechanisms that could slow AI development when necessary for safety.
The apparent contradiction — competing companies arguing for different levels of openness while researchers from those same organizations sign a joint appeal for safeguards — reflects the complexity of the debate. The technology industry broadly agrees that AI poses significant risks, but there is no consensus over how those risks should be managed or who should have the authority to determine acceptable levels of access.
Meta’s own position has evolved as the company has sought to catch up with AI leaders. For years, Zuckerberg championed open-source AI, making Meta’s models broadly available and arguing that developers should be able to experiment with the technology.
That strategy became more complicated as Meta fell behind Anthropic and OpenAI in some areas of the AI race. Zuckerberg subsequently invested $14.3 billion in AI startup ScaleAI and hired its founder, Alexandr Wang, to help overhaul Meta’s AI organization, which was renamed Meta Superintelligence Labs.
Meta spent billions assembling a team of prominent AI researchers around Wang and began developing a more tightly controlled AI model rather than relying exclusively on its earlier open-source strategy.
The shift offered Meta a potentially more lucrative commercial model by allowing the company to charge customers for access to advanced systems. This month, Meta began selling access to its latest AI model, Muse Spark, while continuing to develop open-source models.
The evolution has made Zuckerberg’s criticism of centralized AI power more nuanced. Meta is no longer exclusively committed to giving away all of its most advanced AI technology, even as its chief executive continues to argue that broad access remains fundamental to the company’s philosophy.
Zuckerberg has long predicted that AI will transform society in ways that could ultimately improve people’s lives and expand their capabilities. Meta this month launched a national advertising campaign emphasizing its decision to invest in AI tools and distribute them broadly, describing the strategy as “betting on people.”
He made a similar argument in an opinion article published by The Wall Street Journal this week, criticizing the centralization of AI power and advocating what he described as individual empowerment, invention and a balance of power.
In both the article and the Times interview, Zuckerberg stopped short of directly naming Anthropic and OpenAI. But his criticism comes as those companies increasingly advocate stringent safety measures around increasingly capable AI models.
The disagreement also has a significant geopolitical dimension. The United States is attempting to maintain its lead in AI as China rapidly develops competing models and technologies. American policymakers therefore face a difficult balance between encouraging innovation and preventing AI capabilities from being exploited by adversaries.
For companies such as Meta and Nvidia, open-source development can strengthen the broader American technology ecosystem by allowing more researchers and businesses to work with AI. But critics argue that releasing powerful models too broadly could allow hostile governments or criminal groups to exploit them for cyberattacks, biological research or other harmful purposes.
The debate has therefore moved beyond a conventional Silicon Valley dispute over business models. It increasingly involves questions about national security, technological sovereignty, corporate power and the future distribution of computing capabilities.
Zuckerberg believes the historical trajectory of technology favors his position. In his view, technological tools tend to become more accessible over time rather than remaining permanently controlled by a small group of institutions.
“I believe the general arc of the industry has been toward more openness and putting the power of tech into more people’s hands, not fewer,” he said.
That belief is now being tested by AI, a technology whose capabilities have advanced at extraordinary speed and whose potential risks are still poorly understood.
The central question is whether AI should follow the traditional path of the internet, personal computing and other transformative technologies, with capabilities spreading increasingly widely, or whether its unprecedented potential requires a fundamentally different model in which the most powerful systems remain under tighter institutional control.
Zuckerberg has made clear where he stands. He argues that concentrating AI power could create its own danger by allowing a handful of companies or institutions to determine which values are embedded in technology used by billions of people.
His opponents see the problem differently. For them, the greatest danger may be that advanced AI becomes too powerful before society has developed adequate mechanisms to control it.
As the Trump administration weighs new policies and technology executives lobby Washington, the dispute is likely to become more consequential. The outcome could determine not only how AI companies develop their most advanced systems, but also who gets to use them and how much influence ordinary users, researchers and businesses have over the technology.
For Zuckerberg, the answer is clear: The future of AI should not be defined by a single centralized intelligence or a small group of companies. It should be distributed among individuals, allowing people with different goals and values to shape the technology for themselves.
That vision places Meta at the center of a fight over the fundamental architecture of the AI era — one in which the question is no longer simply how powerful artificial intelligence can become, but who should hold that power.
More in Business coverage
- AI firm halves list of alleged unauthorized secondary markets amid confusion in booming private equity trading.
- The AI company is in talks with U.S. officials after government restrictions prompted global shutdown of its latest models.
- OpenAI has quietly acquired voice-cloning startup Weights.gg as the company deepens investment in artificial intelligence voice technology while navigating growing concerns over copyright, celebrity likeness and digital safety.
- Jensen Huang unveils Nvidia’s first standalone data center CPU as the company expands beyond AI accelerators and challenges Intel and AMD in server infrastructure.
- The Trump administration has established a formal collaboration between leading artificial intelligence developers and operators of essential services to strengthen cybersecurity and address vulnerabilities discovered by advanced AI systems before they can be exploited by malicious actors.
