Inside the $145 Billion Bet, the Benchmark Scandal, and the Employees Caught in the Middle

Meta AI is in the middle of one of the biggest shake-ups in the company’s history, and the signs point to more turbulence ahead.

The company faked benchmark scores on its Llama 4 model, spent $14.3 billion buying a data-labeling startup, pulled thousands of employees off their normal jobs, and got caught tracking staff keystrokes to train its systems.

Each of these events on its own might look small.

Together, they paint a picture of a company scrambling.

👉 Free download: Start a 1-Person Business With Claude AI — Free Quick-Start Guide

This guide walks you through building an income stream that doesn’t depend on any one company’s AI drama, including Meta’s.

If you’ve been following the AI world in 2026, you’ve probably seen the headlines pile up.

Let’s break down exactly what’s going on, one sign at a time.

We strongly recommend that you check out our guide on how to take advantage of AI in today’s passive income economy.

Sign 1: The Llama 4 Benchmark Scandal Exposed a Deeper Problem

In January 2026, Meta’s outgoing chief AI scientist Yann LeCun told the Financial Times something that shook the AI world.

He said Llama 4’s benchmark results were “fudged a little bit.”

LeCun explained that Meta’s team used different versions of the Llama 4 model for different tests.

Then they combined the best scores into one table, making it look like a single model had aced everything.

This is one of the clearest Meta AI shake-up warning signs anyone could ask for.

Normally, AI labs test one version of a model across every benchmark.

Meta didn’t do that with Llama 4, which launched in April 2025 to huge hype and then a wave of disappointment once real users got their hands on it.

The gap between the marketing and the reality was hard to ignore.

CEO Mark Zuckerberg reportedly lost confidence in the team behind the release once the truth came out.

LeCun said Zuckerberg was “really upset” and ended up sidelining the entire generative AI organization at Meta.

That single decision set off a chain reaction that’s still playing out today.

Sign 2: Yann LeCun’s Exit Was Louder Than Most Departures

LeCun didn’t leave quietly.

After more than a decade at Meta, he walked out the door and kept talking.

He told reporters his replacement, Alexandr Wang, “has no research experience and doesn’t know how to do research.”

That’s a striking thing to say about the person now running your former employer’s entire AI division.

LeCun has since started his own company, Advanced Machine Intelligence Labs, focused on building what he calls “world models” instead of chasing bigger language models.

He’s been open that he thinks large language models are a dead end for reaching human-level intelligence, a view that put him at odds with Meta’s entire AI strategy.

This kind of public disagreement from a Turing Award winner is rare.

It’s another one of the Meta AI shake-up warning signs that suggests the turmoil goes beyond one bad product launch.

👉 Get Access to: The AI Blog Monetization Quickstart Guide

If you write about tech news like this for your own blog, this guide shows you how to turn stories like Meta’s AI struggles into steady income.

Sign 3: A $14.3 Billion Deal Installed a 28-Year-Old at the Top

Right after the Llama 4 controversy, Meta didn’t slow down.

Instead, it paid roughly $14.3 billion for a large stake in Scale AI, a data-labeling company.

Meta then installed Scale’s founder, Alexandr Wang, as chief AI officer, putting him in charge of a brand-new division called Meta Superintelligence Labs.

Wang was 28 years old at the time and had never run an AI research team before.

On paper, Yann LeCun reported to him, a strange twist given LeCun’s decades of research experience.

Big acquisitions like this one are meant to fix a problem fast.

But moving that much money and power that quickly can also create new problems, especially inside a research culture that used to run on scientific independence.

This is exactly the kind of decision that turns into one of the clearer Meta AI shake-up warning signs down the line.

Sign 4: Spending Keeps Climbing Even as Doubts Grow

Meta didn’t stop after the Scale AI deal.

In early 2026, the company raised its full-year capital spending guidance for AI infrastructure to a range of $125 billion to $145 billion.

That’s up from an already massive $115 billion to $135 billion estimate set just months earlier.

For comparison, Meta spent about $72 billion on capital expenses in all of 2025.

So the 2026 number could nearly double that.

Wall Street noticed, too.

Meta’s stock dropped sharply after the higher spending guidance was announced, with some analysts openly questioning the return on all that investment.

That kind of investor unease is one of the loudest Meta AI shake-up warning signs so far, because it means even the people funding this bet are starting to worry.

👉 Free download: The Claude AI Digital Product Starter Pack — 10 Done-For-You Prompts for Beginners

While Meta pours billions into AI with mixed results, you can build your own small AI-powered product business using tools that already work well, starting with this free prompt pack.

Sign 5: Thousands of Engineers Got “Drafted” Into Data Labeling

Here’s where things got personal for Meta employees.

Around May 2026, Meta reassigned roughly 6,500 to 7,000 employees into a new group called Applied AI, later referred to internally as Agent Data Optimization.

Many of these were engineers pulled straight off product, infrastructure, and security teams.

Some reports say between 30 and 50 percent of certain core teams were moved into this new group.

Employees didn’t get much say in the matter.

Reports described the choice as simple: join the new unit or leave the company.

Staff started calling themselves “draftees,” and one employee told Wired the job felt “literally the gulag,” spending their days generating puzzles and coding problems just to feed Meta’s AI models fresh training data.

Meta’s own chief product officer, Chris Cox, admitted internally that the environment felt “brutal,” and CTO Andrew Bosworth called the reorganization “atrocious,” with morale reportedly at its lowest point in twenty years.

Zuckerberg later addressed the backlash directly, telling staff on a leaked call that Meta employees have “significantly higher” intelligence than outside contractors, which is part of why he chose to reassign existing staff instead of hiring outside data labelers.

Zuckerberg eventually admitted in a memo that the changes had “caused distress” and promised no further company-wide layoffs for the rest of 2026.

That kind of internal apology, following a decision this disruptive, is another clear entry on the list of Meta AI shake-up warning signs worth watching.

Sign 6: The Keystroke-Tracking Program Made Things Worse

If reassigning thousands of engineers wasn’t enough, Meta then introduced something called the Model Capability Initiative, or MCI, in April 2026.

This software logged employees’ keystrokes, mouse movements, and screen content on their work laptops.

The stated goal was to teach Meta’s AI agents how real people use computers, since Meta wanted training data its competitors couldn’t easily scrape.

When employees asked if they could opt out, CTO Andrew Bosworth reportedly said there was no option to opt out on a company-issued laptop.

More than 1,500 employees signed a petition demanding the program be shut down entirely.

One employee told the BBC the whole thing felt “very dystopian.”

Meta eventually allowed staff to pause the tracking for up to 30 minutes at a time, which many saw as a small concession rather than a real fix.

Then things got worse: an internal security review found that some of the collected data, including private conversations and performance records, had been left exposed across thousands of internal systems, forcing Meta to pause the program entirely while it investigated.

👉 Get Access to the full Package: Start a 1-Person Business With Claude AI

If watching this kind of corporate AI chaos makes you want more control over your own income, this package shows you how to build something independent, powered by tools you actually trust.

Sign 7: Meta AI’s User Growth Has Stalled While Rivals Catch Up

Meta has long pointed to its huge user numbers as proof its AI strategy works.

Back in May 2025, Zuckerberg announced Meta AI had crossed 1 billion monthly active users, a milestone that made headlines at the time.

But more than a year later, Meta hasn’t published a fresh, updated version of that number.

Meanwhile, ChatGPT crossed its own milestone of 1 billion monthly active users in June 2026, and Google’s Gemini app passed 900 million monthly users the same May.

The difference matters because much of Meta AI’s original user count came from people bumping into the assistant inside apps they already use, like Instagram search or WhatsApp, rather than people actively choosing to open a dedicated AI app.

ChatGPT’s growth, by contrast, comes largely from people intentionally going to use it.

A billion people seeing a feature is different from a billion people choosing a product, and that distinction is one of the more overlooked Meta AI shake-up warning signs in this entire story.

Until Meta shares fresh numbers that show real engagement rather than exposure, this sign is likely to stick around.

What This Means Going Into the Rest of 2026

Put these seven signs together and a clear pattern shows up.

Meta is spending more money than almost any company in tech history on AI, while dealing with a benchmark scandal, a high-profile scientist’s public criticism, thousands of unhappy engineers, and a surveillance program that had to be walked back.

None of these problems are fatal on their own.

But stacked together in the same twelve-month window, they suggest Meta’s AI strategy is still being built in real time, not executed from a finished plan.

For anyone building their own business around AI tools, the lesson isn’t to avoid AI.

It’s to build in a way that doesn’t depend on any single company’s internal chaos.

👉 Free download: Start a 1-Person Business With Claude AI — Free Quick-Start Guide

This free guide is a good place to start if you want a simple, steady way to build income using AI tools that are actually stable right now.

And if Meta’s AI shake-up warning signs keep multiplying through the second half of 2026, expect this story to keep making headlines well past this article.

We strongly recommend that you check out our guide on how to take advantage of AI in today’s passive income economy.