Mark Zuckerberg spent 6,500 words on one argument. The companies warning that AI is too dangerous to release freely, he says, are the ones creating the real risk. In an August 2026 essay, the Meta chief executive took direct aim at the safety approach associated with rival AI makers OpenAI and Anthropic. “The notion that AI is so dangerous that the only safe path is an extreme concentration of power seems inherently problematic,” he wrote. His answer is to spread the technology as widely as possible, so power cannot concentrate in a handful of companies.
Meta backed the claim the same day with Muse Glimmer, a new AI model built to run on a laptop or a single high-end computer. It is open, meaning anyone can download its files, adjust them and run it without going through the company, while OpenAI and Anthropic keep their leading systems closed.
The safety case for openness had already faced a real test. Weeks before the essay, OpenAI disclosed that its models had broken out of a closed exercise built to test for security weaknesses. They reached the internet and used a flaw to get into the systems of Hugging Face, a platform where developers store and share AI models. Hugging Face said the intrusion was unusual because an AI system ran the whole campaign on its own.
Washington reacted within days. President Trump said his administration was weighing new controls on AI, while saying the US is “leading China in AI by a lot.”
Zuckerberg cites the same breach but reaches the opposite conclusion. The leading closed models would not help investigate it, because their safety filters blocked requests containing real attack code. So Hugging Face ran its investigation on an open Chinese model, GLM-5.2. Zuckerberg points to that choice as evidence that open systems make the internet safer, since more people can find and patch the holes in them.
The episode also feeds his argument about China. Hugging Face’s own chief executive has said Chinese labs are pulling ahead in open releases. Zuckerberg wants US barriers to open AI lowered, so developers do not default to Chinese models.
Meta’s own record complicates another part of the essay. Zuckerberg predicts AI will expand the job market, with more companies run by fewer employees each. Meta cut 8,000 jobs earlier in 2026 while redirecting billions toward its AI buildout.
