Meta may not have all of its AI models open: Zuckerberg

Meta CEO Mark Zuckerberg has published a detailed letter outlining Meta's future plans for artificial intelligence (AI) technology. In the letter, published yesterday, July 30 (Wednesday), he clearly indicated that not all of Meta's AI models may be open source in the future.

In the letter, Zuckerberg said, "Superintelligence should be open to everyone. But it involves new types of security risks. So which models should be open and which should not be, we need to carefully consider and decide."

This statement is somewhat contrary to Meta's previous position. Until now, Meta has been in favor of keeping its advanced AI models—especially the LLaMA series—open source for everyone. But this time, Meta has clearly stated that some advanced or risky models may no longer be open in the future.

In June 2025, Meta announced that it had begun large-scale work on artificial general intelligence (AGI). As part of this, Meta Superintelligence Labs was formed with an investment of $14.3 billion. Under this unit, more advanced closed AI models are being developed.

The open test of their experimental model ‘Behemoth’ has already been closed, which is a clear message to interested users of Meta—not all models will be open anymore.

The letter also says that future AI will be more personalized. This technology will become a companion to people’s daily lives through devices like smart glasses and VR headsets—understanding the user’s wishes, usage, and thoughts and providing immediate responses.

However, to avoid confusion, a Meta spokesperson clarified, “We are still committed to building strong open models. At the same time, research and work will continue on advanced closed models.”

According to analysts, the future of AI is now not only a matter of technological advancement, but it has also brought a new time to think about security, ethics, and global impact. Meta’s changed position is a reflection of that reality.