Fear and anger brew inside Meta amid AI frenzy

SAN FRANCISCO – A frenzied push for artificial intelligence dominance comes with a different kind of cost for Meta, where massive layoffs, employee surveillance and departures have fuelled reports of a heated internal climate.

As Meta spends billions annually to build out its AI capabilities, employees at Facebook, Instagram and WhatsApp are increasingly unhappy with their Mark Zuckerberg-led parent company.

Meta employees have weathered frequent layoffs since early 2025, including 2026 spring when the company cut 10 per cent of its workforce – some 8,000 jobs – and reshuffled another 7,000 employees.

For those who remain, an internal AI training initiative has drawn accusations of surveillance.

The company also underwent a major reorganisation of its AI research division, into which Zuckerberg, Meta’s founder and chief executive, has poured billions of dollars.

The malaise stands in stark contrast to Meta’s robust finances – driven by advertising, which makes up nearly 98 per cent of its revenue. In the first three months of 2026, Meta’s net income rose to more than US$26 billion (S$34 billion).

However, the bill for its AI investments is also exploding, prompting Zuckerberg, who has near-absolute power over the company, to impose sweeping cuts and increase monitoring of employees in the name of efficiency and savings.

The cuts are funding a massive race for infrastructure: Meta plans to spend up to US$145 billion on AI investments in 2026, nearly twice that of 2025.

After thousands of employees were reassigned to Meta’s AI division, some, speaking anonymously to US media, have complained of “mind-numbing” tasks designed to train machines or even automate away their own jobs.

That controversial programme, called the Model Capability Initiative, was rolled out in April and suspended on June 22. It captured clicks, keystrokes and browsing activity of US employees to train AI agents – software capable of independently performing tasks.

Zuckerberg, who has made AI the company’s north star, defended the programme during an internal meeting: “AI models learn by watching really smart people do things,” he said, according to Wired magazine.

But the tool sparked a revolt. More than 1,600 employees signed a petition calling for it to end, with some likening the company to a “data extraction factory”, according to media reports.

The pause came after private conversations and performance data inadvertently became accessible to all staff. The system risked drawing the attention of European regulators, since it captured exchanges between employees on both continents.

In a statement to AFP on June 30, a Meta spokesperson said the programme was designed with privacy safeguards.

“While we have no indication at this time that any data was improperly accessed by Meta employees, we’re pausing it while we investigate,” the statement said.

One employee summed up the mood with a meme from the TV series The Office, posted on an internal company forum, reading: “0 days since our last nonsense.”

All of these efforts aim to make up for a persistent lag behind Google, OpenAI and Anthropic, which dominate the race for cutting-edge AI models. Meta’s own models, repeatedly delayed, have proved disappointing even internally.

To regain ground, Zuckerberg invested over US$14 billion in 2025 into Scale AI, a San Francisco-based start-up, and poached its CEO Alexandr Wang – who was 28 years old at the time – to run a “superintelligence” lab inside Meta.

The expensive bet has yet to win people over. Several key figures have since walked out, among them Yann LeCun, considered one of the “godfathers” of modern AI, who had led Meta’s AI research since 2013.

LeCun suddenly found himself reporting to Wang, more than 35 years his junior. He left Meta at the end of 2025 to launch his own start-up.

In an interview with the Financial Times, the Turing Award winner lamented that, although “he learns fast”, Wang has “no experience with research” and was on “a dead end” quest.

The stakes for Meta go beyond its social networks now. The company is also doubling down on consumer electronics with smart glasses and is considering a new prediction-market app called Arena, potentially in partnership with Polymarket and Kalshi, according to The New York Times.

Lawsuits also threaten to consume time and resources.

For the first time, a Los Angeles jury in March found Meta liable for the effects of social media addiction, just one day after a separate ruling in New Mexico said Meta had failed to protect minors.

Meta has appealed, but more lawsuits are expected in 2026. AFP