Meta's Shift: Zuckerberg Signals Potential Retreat from Open-Source AI in Pursuit of 'Superintelligence'

Zuckerberg signals Meta won't open source all of its 'superintelligence' AI models | TechCrunch

Meta CEO Mark Zuckerberg shared his vision on Wednesday for "personal superintelligence," the idea that people should be able to use AI to achieve their personal goals. Smuggled into the letter is a signal that Meta is shifting how it plans to release AI models as it pursues "superintelligence." "We believe the benefits of superintelligence should be shared with the world as broadly as possible," wrote Zuckerberg. "That said, superintelligence will raise novel safety concerns. We'll need to be rigorous about mitigating these risks and careful about what we choose to open source." That wording about open source is significant. Zuckerberg has historically positioned Meta's Llama family of open models as the company's key differentiator from competitors like OpenAI, xAI, and Google DeepMind. Meta's goal has been to create open AI models that were as good as or better than those closed models. In a 2024 letter, Zuckerberg wrote, "Starting next year, we expect future Llama models to become the most advanced in the industry." And while many say Llama doesn't fit the strict definition of open source AI - partly because Meta hasn't released its massive training datasets - Zuckerberg's words point to a possible change in priority: open source may no longer be the default for Meta's cutting-edge AI. There's a reason why Meta's rivals keep their models closed. Closed models give companies more control over monetizing their products. Zuckerberg pointed out last year that Meta's business isn't reliant on selling access to AI models, so "releasing Llama doesn't undercut our revenue, sustainability, or ability to invest in research like it does for closed providers." Meta, of course, makes most of its money from selling internet advertising. Still, that stated viewpoint on open models was before Meta started to feel like it was falling behind competitors, and executives became obsessed with beating OpenAI's GPT-4 model while developing Llama 3. Cut to June 2025, when Meta began its public AGI sprint in earnest by investing $14.3 billion in Scale AI, acquiring Scale's founder and CEO, and restructuring its AI efforts under a new unit called Meta Superintelligence Labs. Meta has spent billions of dollars to acquire researchers and engineers from top AI firms and build out new data centers. Recent reports indicate that all that investment has led Meta to pause testing on its latest Llama model, Behemoth, and instead focus efforts on developing a closed model. With Zuckerberg's mission for introducing "personal superintelligence" to the world - a decided shift from the rivals he says are working on "automating all valuable work" - his AI monetization strategy is taking shape. It's clear from Zuckerberg's words today that Meta plans to deliver "personal superintelligence" through its own products like augmented reality glasses and virtual reality headsets. "Personal devices like glasses that understand our context because they can see what we see, hear what we hear, and interact with us throughout the day will become our primary computing devices," Zuckerberg wrote in Wednesday's letter. When asked about Meta potentially keeping its most advanced models closed, a Meta spokesperson said that the company remains committed to open source AI and said it also expects to train closed-source models in the future. "Our position on open source AI is unchanged," a spokesperson said. "We plan to continue releasing leading open source models. We haven't released everything we've developed historically and we expect to continue training a mix of open and closed models going forward."

Zuckerberg Walks Back Open-Source AI Pledge, Citing Safety Risk

Meta CEO Mark Zuckerberg is hyping up the emergence of superintelligent AI. Just don't expect it to be open source, like the company's Llama large language models. On Wednesday, Zuckerberg published a manifesto about the company's efforts to build AI that's smarter than humans. "Developing superintelligence is now in sight," he writes. "Over the last few months, we have begun to see glimpses of our AI systems improving themselves. The improvement is slow for now, but undeniable," he said. But while the manifesto talks up Zuckerberg's vision to use AI to improve society, he also signals a retreat from openness, citing safety. "We believe the benefits of superintelligence should be shared with the world as broadly as possible. That said, superintelligence will raise novel safety concerns. We'll need to be rigorous about mitigating these risks and careful about what we choose to open source," he says. The statement stands out since it was only a year ago that Zuckerberg wrote a separate article pledging to make the company's AI open source. "There is an ongoing debate about the safety of open source AI models, and my view is that open source AI will be safer than the alternatives," he said at the time. Other critics say Zuckerberg's AI efforts were never really open source, which is supposed to mean that a software's code is freely available for developers to inspect and modify. But with Llama, Meta has imposed certain restrictions, according to the Open Source Initiative, a group that stewards the open-source definition. For example, users need to sign off on Meta's own licensing document to download Llama. Although the agreement is designed to prevent malicious use of the AI technology, it also prohibits deploying Llama in critical infrastructure or for regulated controlled substances. Meta's agreement also requires users to seek out a license if the Llama implementation gains "greater than 700 million monthly active users in a calendar month." In addition, the company has tightly restricted Llama's use in the European Union and has not released the training data to build the large language model. Of course, keeping Meta's superintelligent AI under wraps is also about staying ahead of the competition. Zuckerberg has been offering massive deals to poach top AI researchers from rivals like OpenAI and Apple. At the same time, Meta is investing billions in new data centers that promise to develop and run future AI systems. In his manifesto, Zuckerberg addd that his intent it to deliver the AI capabilities to "billions of people across our products."

Is Mark Zuckerberg flip flopping on open source AI?

Meta's CEO has suggested his AI "superintelligence" might take a different approach. Earlier today, Mark Zuckerberg shared a rambling outlining his vision to build AI "superintelligence." In the memo, Zuckerberg hinted that the pursuit of more powerful AI might require the company to be more selective in what it open sources. Citing "safety concerns" he wrote that Meta would need to be "rigorous" about such decisions. The line stood out to many as Zuckerberg -- who once said "" in reference to closed platforms -- has made open source central to Meta's AI strategy. During Meta's second quarter earnings call, Zuckerberg further acknowledged there could be a shift, though he downplayed the significance of it. Here's what he said when asked if his thinking had changed. That's notably different than what he wrote almost exactly a year ago in a titled "Open Source AI is the Path Forward." In that, even longer note, he said that open source is crucial for both Meta and developers. "People often ask if I'm worried about giving up a technical advantage by open sourcing Llama, but I think this misses the big picture," he wrote. "I expect AI development will continue to be very competitive, which means that open sourcing any given model isn't giving away a massive advantage over the next best models at that point in time." He also argued that open source is safer. "There is an ongoing debate about the safety of open source AI models, and my view is that open source AI will be safer than the alternatives. As long as everyone has access to similar generations of models - which open source promotes - then governments and institutions with more compute resources will be able to check bad actors with less compute." To be clear, Zuckerberg said the company would continue to open source some of its work. But he seems to be laying the groundwork for a future in which Meta's "superintelligence" could be a lot less open.