Mark Zuckerberg has long been vocal about keeping Meta’s AI models open to the public.
But a recent message from Meta CEO suggests that policy might be shifting, especially as the company chases more advanced, potentially risky forms of artificial intelligence.
In a July 2025 update, Zuckerberg introduced the idea of “personal superintelligence”: AI tools designed to help individuals reach personal goals, not just automate jobs.
But buried in the note was a subtle pivot: future models may not be as open as before.
What Did Zuckerberg Actually Say?
Zuckerberg wrote that while Meta wants the benefits of superintelligence to be widely shared, “we’ll need to be rigorous about mitigating these risks and careful about what we choose to open source.”
That sounds like a polite way of saying: some of the really powerful stuff might stay behind closed doors.
And for a company that once made “openness” a core part of its AI strategy, that’s a big deal.
From Open-First to Open-Maybe?
Meta’s previous approach stood in contrast to competitors like OpenAI and Google DeepMind, which tend to keep their most advanced models locked down.
Meta’s Llama models, especially Llama 2 and Llama 3, were released openly, sort of.
They weren’t fully “open source” by strict definitions, since the company didn’t share training data or full development pipelines.
Still, it was more open than others.
Zuckerberg even said in 2024 that future Llama models would be the industry’s most advanced and free to use. But now, it seems like that pledge came with a clause.
“If at some point… we feel like it’s not responsible to open source it, then we won’t,” he said on a podcast last year.
So, Meta left the door open for this shift all along.
Why Close the Door Now?
1. Pressure to Keep Up with OpenAI
Court documents and insider reports show Meta executives have been laser-focused on catching up, or beating, OpenAI’s GPT-4. The competitive heat has clearly intensified.
2. Big Investments, Big Expectations
Meta recently invested a whopping $14.3 billion in Scale AI, hired the company’s founder, Alexandr Wang, and created a new internal group called Meta Superintelligence Labs. That’s not small money.
They’ve also been hiring top researchers from OpenAI and other AI labs, and building massive new data centers to support future models.
3. Pivoting Away from Llama
Instead of continuing open tests of its next model, called Behemoth, reports suggest Meta has paused testing. Now, the focus is on developing a closed model, one that Meta controls tightly.
Personal Superintelligence: Meta’s New Focus
So what is Zuckerberg after?
Instead of building AIs that replace jobs, Meta wants to create smart assistants that live in your personal tech, especially AR glasses and VR headsets.
Imagine glasses that:
- Understand what you see and hear
- Help you remember things
- Give suggestions in real time
- Learn from your habits throughout the day
That’s the future Zuckerberg is selling: an AI that feels more like a personal sidekick than a job replacement.
“Personal devices like glasses… will become our primary computing devices,” he wrote.
So, Is Meta Still Open Sourcing AI?
Kind of, but also kind of not.
Here’s what Meta told reporters:
“We plan to continue releasing leading open source models… but we expect to train a mix of open and closed models going forward.”
In other words:
- Some tools will still be open
- Others, especially more powerful ones, may stay private
It’s not a total reversal. But it’s definitely not the same “open by default” vibe they pushed before.
Why the Change Matters
Keeping models closed isn’t just about safety, it’s also smart business.
Closed models:
- Offer more control
- Are easier to monetize
- Help avoid PR disasters from misuse
Open models:
- Boost innovation
- Get more feedback and testing
- Allow developers and smaller companies to build on top
Meta says their business doesn’t rely on AI sales, since most of their revenue still comes from ads. But with billions spent on AI infrastructure, that could change fast.
Final Thoughts
If you’re just someone who enjoys using AI tools, chatbots, writing assistants, and voice helpers, this shift could mean:
- Some tools stay free and open
- Others are only built into Meta devices
- Advanced features may require subscriptions or hardware
It also means Meta’s AI won’t be as easy to build on for developers, startups, or students. If Meta closes up, fewer people will be able to peek under the hood.
But on the flip side, this may also mean safer tools, better user experiences, and faster progress toward truly useful personal AI.
Quick Recap
| What’s Changing | Details |
| Open-source strategy | Becoming more selective |
| Llama project | Testing paused; may be replaced by closed models |
| AI focus | From general automation to personal superintelligence |
| Investment | Over $14 billion into AI infrastructure and talent |
| Device push | Emphasis on AR/VR wearables to deliver AI experiences |
What to Watch Next
- Will Meta release another Llama model in the open?
- How will Meta’s AI stack up against GPT-5?
- Could this signal a trend away from open AI across the industry?
Only time will tell.
But one thing’s clear: the AI race is heating up, and Meta just made a big move.
