Meta CEO Mark Zuckerberg is proposing that leading AI labs give the U.S. government an early look at their most powerful models, before those systems are finished or publicly released, to ease security fears without slowing the industry down.
The idea came tucked inside a sweeping, roughly 6,500-word essay Zuckerberg published Monday, titled "The Future is for Everyone," that doubles as a manifesto for spreading artificial intelligence as widely as possible.
"Rather than waiting until a model is ready to release for the government to review and start using it," Zuckerberg wrote, "leading labs should provide the government with intermediate training checkpoints of new advanced models and technical staff so the government can harden and secure critical systems."
Such a move would avoid "restricting or delaying individuals’ access to personal superintelligence or causing an imbalance of power."
Under Zuckerberg's plan, the so-called frontier labs building the most advanced AI would share their new programs with the government for review while they are still being developed, and also provide technical staff, so agencies can harden critical systems. He framed it as a way to give Washington a security edge without forcing labs to hold products back, because even a one-month release delay, he argues, could let foreign rivals race ahead.
The pronouncement might carry extra weight because Meta has historically been one of the most prolific producers of open-source AI models, which can be easily downloaded, inspected and run by users with the right setup or technical know-how.
Zuckerberg's post came the same day Meta released a new open-source AI model, Muse Glimmer, which the company said is optimized to run on users’ local computers, "enabling use cases that range from local agents and function calling, to local coding."
The debate over whether AI models should be open-source, and whether the new systems pose such dire national-security or other risks that they should be closely guarded by government overseers, has ripped through Silicon Valley and Washington over the past month.
Open-source AI models
More than 75 companies signed Nvidia CEO Jensen Huang's recent letter pressing the Trump administration to back open-source AI, DeAI News reported. U.S. President Donald Trump's administration, previously seen as an ally of the AI industry, has delayed model releases by Anthropic, while also accusing Chinese AI companies of ripping off American designs.
Anthropic, whose powerful Claude AI models are not typically open-source, was a notable holdout from the coordinated effort.
Zuckerberg's call to keep shipping powerful models comes just days after OpenAI hit pause on its unreleased Astra system, saying it could not rule out that the model was capable enough to help mount serious cyberattacks, the very fear that skeptics of open models cite.
The Facebook founder warned that the "most dangerous scenario" would be labs "training powerful models and keeping them for themselves."
(HOW AI WAS USED IN THE PRODUCTION OF THIS PIECE: The first draft of this story was written on Claude Desktop using a custom blog-rewrite skill plus background research. I used our Distro Publisher MCP connector to file the story to DistroVerse, our web app, where I used the built-in (non-AI) story-editing interface to edit the piece. The word count was tallied directly from the post, and facts were checked against Zuckerberg's original essay, OpenAI's cyber-capabilities post, and prior DeAI News reporting before publication.)