Mark Zuckerberg criticizes closed AI rivals as Meta returns to open models

Mark Zuckerberg wants you to know that locking up AI behind corporate walls is not just bad business. It’s bad for America.

In an interview published July 28, Meta’s CEO took direct aim at companies like OpenAI and Anthropic, arguing that their closed-door approach to AI development risks concentrating too much power in too few hands. The timing was deliberate: Zuckerberg used the moment to signal that Meta is pivoting back toward open-source model releases after a brief and somewhat awkward detour into proprietary territory.

The proprietary detour that didn’t quite land

In April 2026, Meta released Muse Spark, the company’s first major proprietary AI model. It was a notable departure from the Llama series, which Meta had built its AI reputation on by releasing openly for developers and researchers to use, modify, and build upon.

Meta’s earlier Llama models received a somewhat underwhelming reception relative to the frontier capabilities being demonstrated by competitors working behind closed doors. Muse Spark was Meta’s attempt to play that game.

During discussions around Meta’s Q2 2026 earnings, Zuckerberg indicated the company plans to resume releasing open-source models in the near future.

The philosophical battle lines

Zuckerberg’s critique of closed AI labs was not subtle. He framed openness as an extension of American values, contrasting it with what he characterized as a doom-focused mentality at rival labs that use safety concerns to justify keeping their most powerful models locked down.

Where the industry is lining up

Zuckerberg isn’t fighting this battle alone. Microsoft, Nvidia, and Google have all shown varying degrees of alignment with Meta’s open approach, though each company calibrates its own strategy differently.

The strategic calculus for Meta is also simpler than it looks. Meta doesn’t primarily monetize AI models directly. It monetizes attention across its family of apps. OpenAI and Anthropic, by contrast, sell model access as their core business. Their incentive to keep models closed is structural, not just philosophical.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.