Meta’s AI ambitions: hiring revolutionaries into a tech bureaucracy

After poaching staff from OpenAI and elsewhere, Mark Zuckerberg has a new plaything – Meta Superintelligence Labs. The tech giant obviously has loads of cash to do it, but will it succeed?

It all moved very quickly once OpenAI CEO Sam Altman said in mid-June that Meta was offering his employees huge signing bonuses of $100 million. He sounded confident that “none of the best people” would leave.

This wasn’t to be. Soon, Meta indeed poached three OpenAI researchers – Lucas Beyer, Alexander Kolesnikov, and Ziaohua Zhai – in a recruiting coup to join Zuckerberg’s company. A couple of days later, four more AI researchers from OpenAI followed suit.

Finally, Zuckerberg announced a major restructuring of Meta’s AI group, calling it Meta Superintelligence Labs in an internal memo sent out to employees on Monday.

Zuckerberg’s sky-high ambition

Among other things, the new group will focus on developing AI superintelligence, a system that can complete tasks as well as or better than humans, Bloomberg said, adding that the company is planning to spend hundreds of billions of dollars on AI research in the years to come.

The Superintelligence Labs will be led by Alexandr Wang, former CEO of data-labeling startup Scale AI, and Nat Friedman, former CEO of Github, Zuckerberg wrote in the internal message.

“As the pace of AI progress accelerates, developing superintelligence is coming into sight. I believe this will be the beginning of a new era for humanity, and I am fully committed to doing what it takes for Meta to lead the way,” said Meta CEO.

Clearly, AI has become Zuckerberg’s top focus this year as he’s eager not to lose the race to rivals such as OpenAI and Google. He has reportedly personally worked to recruit talent for the new teams.

Ambitions seem sky-high. However, while some say Meta’s rise to the top is nearly inevitable due to the sheer depth of the company’s pockets, others have doubts about whether Zuckerberg will succeed, saying Meta’s business model is simply not fit for any meaningful AI innovation.

According to some, there’s something within Meta’s culture that could block potentially game-changing developments. One expert told Cybernews that Meta was essentially “hiring revolutionaries into a bureaucracy.”

Besides, with Meta’s history of abusing user data and their privacy in mind, how confident can we really be that the company won’t be exploiting our data to train their large language models?

Deep pockets, massive resources

Jacob Anderson, CEO of Beyond Ordinary Software Solutions, is an optimist. To him, Meta is already a unique player in the field of AI because Meta AI, powered by the new Llama 4 deep learning model, is different from other standalone AI applications.

For example, the app can act as a companion to Meta’s Ray-Ban AI glasses, enabling hands-free interaction and integration with wearable technology.

“Meta is the only frontier model that you can put on your body and have it interact with the real world in real time. There is no other device like it available today. Meta is clearly the forerunner of AI-enhanced Reality,” Anderson told Cybernews.

Syed Hussain, CEO and founder of AI startup SHIZA, agrees. He thinks that Meta has already changed the game and will dominate AI because it open-sourced Llama and “turned the entire industry into its unpaid research and development team.”

“Meta’s acquisition genius isn't just hiring talent, it's preserving research cultures while providing massive resources, like they did with Instagram and WhatsApp,” said Hussain.

More generally speaking, Meta does have a strategic advantage, said Wyatt Mayham, CEO of Northwest AI Consulting.

“Meta’s acquisition genius isn't just hiring talent, it's preserving research cultures while providing massive resources, like they did with Instagram and WhatsApp,”

Syed Hussain

“Meta has unique assets, which are its massive computing resources and obviously never-ending pockets to poach top talent from OpenAI and Google. Being open-source also helps position themselves differently from their closed competitors,” Mayham told Cybernews.

Zuckerberg’s previous object of obsession, the metaverse project, is lagging and losing Meta billions of dollars every year. Why is the metaverse’s struggle not a sign that the company’s bet on generative AI could end similarly?

To Hussain, the metaverse wasn’t a failure because it’s too early to say: “AI-driven unemployment will create massive demand for virtual experiences just as AI makes them compelling.”

Is Meta even good at innovation?

Others find the metaverse's example telling, especially since Zuckerberg, like last time, is planning to invest billions into Meta’s new AI-powered adventure.

“It seems like they are at risk of repeating their metaverse pattern, which is throwing massive resources at a problem without the execution culture needed for innovation,” said Mayham.

That might be the crux of the matter. You can afford to hire elite talent and build multiple data centers, but compute and capital don’t equal breakthrough cognition in the context of building superintelligence.

“What separates true superintelligence is adversarial iteration, autonomy, the ability to self-modify in unpredictable environments,” Nic Adams, co-founder and CEO of 0rcus, a cybersecurity company, told Cybernews.

“None of that survives inside a public company optimized for ad revenue, optics, and risk management.”

“You can poach ex-OpenAI or DeepMind researchers all day, but if you drop them into a system that prioritizes quarterly milestones, internal consensus, and brand safety, their edge gets blunted fast,”

Nic Adams.

Which is, of course, what Meta is, Adams adds, because its core operating DNA is built on metric-driven iteration and behavioral manipulation instead of discovery.

“You can poach ex-OpenAI or DeepMind researchers all day, but if you drop them into a system that prioritizes quarterly milestones, internal consensus, and brand safety, their edge gets blunted fast,” he says.

“The companies that win the AI arms race will embrace dissonance, risk, and technological offense. Meta is trying to hire revolutionaries into a bureaucracy. Unless they spin up a fully insulated rogue lab, culture alone will neutralize the upside.”

Bryan Woolgar-O'Neil, chief technology officer and co-founder at Harmonic Security, another cybersecurity firm, is of the same view. He says there’s plenty of reason to doubt the long-term success of Meta’s move.

According to him, Meta still lacks meaningful enterprise distribution, which is the glue holding together the AI strategies of rivals like Microsoft and Google. They’re already embedded in the workflows, the APIs, and the procurement habits of global business.

“Whether they like it or not, Meta is still the company of social media and VR headsets. There’s also a cultural side to it: moving fast and breaking things works fine when you’re shipping social features, but it’s borderline reckless when you're developing systems that shape how billions interact with truth, work, and creativity,” Woolgar-O'Neil told Cybernews.

The user data and privacy problem

Even the release of Llama 4 in April had major hiccups. While Meta proudly called the AI system “one of the smartest in the world,” rumours quickly spread on the web that the model is actually struggling to deliver performance despite strong benchmark claims.

“This has really damaged their credibility, just as competitors like DeepSeek released their models at a fraction of the cost, and it was significantly better than Llama 4,” Mayham told Cybernews.

“The Llama 4 problems seem like deeper organizational and cultural issues within their GenAI division. Their head of AI research just left before the failed launch, which certainly points towards internal dysfunction.”

And then there’s the question of respect, or lack of it, for user privacy. According to a recent report by Incogni, Meta AI has been ranked as the worst AI platform for data privacy among nine platforms assessed.

“It’s likely that users’ personal data is already used in the training data for these models, whether derived from social media interactions or not,” said Incogni. “Currently, there’s no way for users to opt out of their data being used to train these models.”

In late May, despite heavy activist criticism, Meta already began training its AI models using the public posts of European Instagram and Facebook users.

“Meta’s past is a clear indication of their priorities. Time and time again, the company has shown that user data is its primary revenue stream, not a trust role,” Aaron Whittaker, VP of demand generation and marketing at Thrive, a marketing agency, says.

“Its business model is at its heart based on data collection and data analysis, and thus a development of privacy-respecting AI is economically contradictory to their fundamental strategic line.”

Adams agrees: “Meta’s entire infrastructure is engineered around psychographic extraction and behavioral forecasting. The large language models just give them a more powerful compression algorithm to repurpose attention at scale.”

“Whether it’s message metadata, camera vectors, or attention dwell-time, Meta will inevitably use every legal and gray-area method to feed its models,” said Adams.

The only hope, according to Anderson, is that Meta has already felt the sting of backlash with regard to misuse of user data and is now very cautious. Still, it also means that the company is also much more careful about how we’re made aware of what they’re doing.