Zuckerberg Argues Against a Few Firms Owning AI

Mark Zuckerberg published a long essay this week laying out why he thinks the future of artificial intelligence should be open, and why he believes the alternative is dangerous. The piece ran to about 6,500 words, and its argument was pointed. The idea that AI is so risky that only a handful of players can be trusted to control it, Zuckerberg wrote, is the thing we should actually worry about.

"The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic."

That line reads as a shot at OpenAI and Anthropic, the two companies most associated with keeping their most capable models closed. Zuckerberg's warning went further. If superintelligence ends up controlled by a small set of people, companies, governments, or the systems themselves, he argued, the results will tilt against everyone left outside that circle.

The essay came with a product

The timing was deliberate. Meta released the argument alongside Muse Glimmer, a lightweight open model that runs on a personal computer, under a permissive license. The company also said developers would get a route to a stronger model, Muse Spark 1.2. So the manifesto is not just philosophy. It is the case for a strategy Meta is already shipping.

Zuckerberg's sunnier claims drew eye-rolls. He sketched a world where anyone could start a business, get tutoring at a doctoral level, or receive tailored lifestyle advice, all from freely available AI. Critics called that vision fanciful, and they have a point about the gap between the pitch and what today's models actually deliver.

An old debate with higher stakes

There is also a tension inside Meta's own record. Zuckerberg has said before that the company needs to be careful about what it chooses to open-source once systems approach superintelligence, which sits awkwardly next to this week's full-throated defense of openness. The honest read is that even the loudest advocate for open AI wants some limits, and the real argument is about where those limits fall.

The open-versus-closed question keeps getting sharper as the models get more capable. It runs straight into national security, where some officials worry that open weights hand rivals a ready-made tool. Zuckerberg's essay is the clearest statement yet of the other side of that fight, from the person with the most to gain if it wins.