Meta's chief executive is making a philosophical case for open artificial intelligence, but the company's recent history of overstated benchmarks and strategic reversals raises questions about whether it can credibly champion that vision. Mark Zuckerberg published an op-ed in the Wall Street Journal arguing that superintelligence should not remain locked inside a handful of labs, but instead be distributed so ordinary people, small businesses, and entire economies can access tools currently reserved for well-funded corporations.

The argument itself has merit. History shows that transformative technologies like aviation, electricity, and personal computing only became truly world-changing once they moved beyond their original creators and became accessible to ordinary developers and small businesses. Meta's decision to release Llama as an open-weight model, despite technical limitations, has genuinely lowered barriers for developers in countries like Kenya who could never afford access to closed frontier models through an application programming interface (API), let alone the computing power needed to train one from scratch.

What Happened to Meta's AI Leadership in 2025?

The credibility problem emerges when examining what Meta actually did during 2025. When Llama 4 launched in April 2025, independent testers could not reproduce Meta's benchmark results. Researchers discovered that the model submitted to a public leaderboard was an unreleased experimental chat version rather than the one available to users, a discrepancy many in the AI research community viewed as an attempt to artificially improve the model's standing.

The situation worsened when Yann LeCun, Meta's outgoing chief AI scientist, told the Financial Times in January 2026 that the results had been "fudged a little bit". This controversy triggered a company-wide AI reorganization, Meta's investment of more than $14 billion in Scale AI, and a gradual shift away from fully open-weight models toward a more closed approach under Meta Superintelligence Labs.

"The results had been fudged a little bit," stated Yann LeCun, Meta's outgoing chief AI scientist.

Yann LeCun, Chief AI Scientist at Meta

This trajectory undermines Zuckerberg's central argument. The case for distributed superintelligence only works as a safety argument if the company promoting it is honest about what it has built. A company that overstated its own benchmark results just months before publishing an essay about transparency is not automatically disqualified from making that argument, but it is also not the impartial voice it presents itself to be.

Does Meta's Business Model Match Its Open AI Philosophy?

A deeper contradiction exists in Zuckerberg's positioning. He built Meta by concentrating attention, user data, and advertising revenue more aggressively than almost any other company. Facebook, Instagram, and WhatsApp are among the most centralized digital platforms ever created, and Meta has spent the past two decades defending that model in courts and before lawmakers.

Now Meta argues that concentration is one of the biggest risks of the artificial intelligence era. Coming from a company that built its success on concentrating so much power, that argument deserves scrutiny. It is less about outright hypocrisy and more about a company that benefited from one kind of gatekeeping while warning against another that it does not control.

How to Evaluate Meta's AI Strategy and Claims

• Examine Track Record: Look at whether Meta's actual model performance matches its public claims and benchmark results, not just the company's stated philosophy about openness.
• Compare Business Incentives: Consider that Meta's emphasis on open AI may reflect its competitive position relative to OpenAI and Google, rather than purely altruistic motivations.
• Assess Capital Commitment: Evaluate whether Meta's projected capital spending of up to $135 billion in 2026 translates into models that earn influence through performance, not just through spending power.
• Monitor Strategic Shifts: Track whether Meta continues moving toward closed, internal models under Meta Superintelligence Labs, which would contradict its open AI messaging.

Zuckerberg is not presenting himself as a neutral AI safety researcher like some teams at Anthropic and OpenAI. Instead, he is positioning Meta as the open alternative to a future where artificial intelligence is controlled by a few companies. While this is framed as benefiting the public, it also supports Meta's business interests.

That approach fits Meta's position. The company cannot match the spending power of OpenAI and Google in the race to build closed AI models the way it once outspent competitors in social media. By making openness its defining advantage, Meta is competing on terms that better suit its strengths.

Zuckerberg's record does show some strength that critics often overlook. Reality Labs lost tens of billions of dollars, and Horizon Worlds never gained widespread adoption. However, the data centers, custom chips, and talent Meta built during that period became the foundation for its rapid shift into artificial intelligence. Few executives could absorb losses on that scale and reuse those investments for a second major push so quickly.

Llama's setbacks and Meta's move toward more closed, internal models suggest its position as an open-source champion is becoming harder to support. Zuckerberg has the financial resources and global reach to keep influencing the direction of the artificial intelligence conversation. What he still needs to prove is that Meta can build a model that earns that influence through its performance, not simply through the amount of money the company spends.

When it comes to trusting Zuckerberg's judgment on artificial intelligence, he has earned a place in the conversation, but he has not yet earned the role of referee. His argument about concentration in artificial intelligence deserves attention, but it deserves scrutiny even more.