Meta Reassigns 7,000 to AI Teams

Recently, Meta CEO Mark Zuckerberg laid off 8,000 employees to focus on a broader AI-focused restructuring, while 7,000 employees were reassigned to new AI teams to strengthen the company’s position in the AI sector.

Business Insider reviewed an internal memo sent to thousands of employees informing them they were selected to be a part of a new team for the new Meta strategy.

Meta Email Announces Reassignments

The emails reviewed by Business Insider state, “This is a reflection of your impact. You were identified as someone who can make a real impact on this team.”

These employees will function within AI-focused teams under the leadership of engineering vice president Maher Saba and report to technology officer Andrew Bosworth. Other teams will focus on AI agents and model development in the “Agent Transformation Accelerator and Agent Data and Optimization.”

Meta has already established a superintelligence lab, reorganized smaller AI-focused teams, and modified some job titles to “AI builder.”

Employee Reactions

Some employees have discussed the changes on internal forums and workplace apps. They range from relief to uncertainty, and some described the process as an “AI draft.”

One employee wrote on Discord, “I got drafted.” Another announced, “Welcome to the draft.”

The “drafted” employees who spoke with Business Insider said the initiative appears to be mandatory. These employees have expressed uncertainty about the new endeavor. Some believe the new AI task forces will focus on data labeling and model training.

This aligns with the expertise of Meta AI chief Alexandr Wang, who previously led Scale AI before joining Meta.

Other AI Initiatives at Meta

The company launched the Model Capability Initiative earlier this year. It is a tool that tracks employee interactions with computers to aid in the training of AI systems using real-world examples of human workflows.

The software was announced with an internal post that read, “For agents to understand how people actually complete everyday tasks using computers, we need to train our models on real examples.”

A recording of an internal meeting was leaked. In the meeting, Zuckerberg focused on Wang’s data labeling experience and suggested the company could gain an advantage using its own workforce to train AI systems.

Zuckerberg said, “I think that this is going to be a very big advantage if we can do it.” He added the idea was a “hypothesis” at the time.

Meta AI Clickbait

In April 2025, the Meta AI app was first launched with a public “Discover” feed that showed AI-generated images and conversations from other users, who seemed unaware they were made public. Now, the app has a standard chatbot interface and a For You page that displays suggested articles that, when tapped, generate entire “stories.”

A London-based reporter for The Verge, Robert Hart, says the prompts for him were aggressively British, involving topics like manners, tea, royals, pubs, soccer, and the art of queuing. The suggested stories included “The Anatomy of the Devastating British Tut,” “A Royal Butler Finally Settled the Milk First Debate,” “The Psychology  of joining a queue Without Knowing Why,” and “Inside the Extreme Sport of Visiting Every UK Pub.”

Meanwhile, the journalist’s colleague was placed within the “watch aficionado bracket” by the algorithm. His feed offered these titles: “My Fake Rolex Experiment” and “The Brutal Math Behind the Rolex Waitlist Illusion.”

The non-sourced, AI-generated stories read like “puffy filler” with little substance beyond restating the prompt’s premise.

Hart sought out the origin of these “stories.” He traced the royal butler tea story to a 2018 BBC Three Comedy series called “Miss Holland” that follows a fictional beauty queen from a small Dutch town. She travels to Britain to learn “how to be posh and classy” from a former royal butler named Grant Harrold.

“The Rolex Experiment” seemed to be a complete fabrication, generated in the chat box as a first-person narrative without a byline. Hart says other stories gave vague references to unnamed experts or fictional research.

He says when he clicked on the same story more than once, the generated fable remained within the rough bounds of the prompt as clear versions of the same thing but were slightly different.

When Hart typed out the same headline into a separate chat, it produced an entirely different response. His chat history was the biggest giveaway, as it revealed the suggested prompts that triggered the generated stories. One of these prompts began:

“You are a helpful conversational assistant. The user is responding to a proactive feed card that was shown to them. The card context below provides background on what prompted the user’s message.” This prompt was followed by references to internal information, instructions, and metadata.

Images were attached to the stories. Many of them Hart described as “bland mush of cartoony people, landscapes, and food.” However, he says that some of the pictures depicted real people, public figures, riddled with errors. “Who really pays for the royal family in 2026?” featured two images of Queen Elizabeth II even though she passed away several years ago and was only one person.

The Queen clones were surrounded by people who were approximations of other royals: a Princess Kate-ish face, an odd attempt at Prince William, and a King Charles-like figure who was an exaggerated resemblance to his late father. There were other usual AI tells like strange/impossible hands and bodies in unnatural angles. One image was a GIF of an older couple dancing and moving their arms in a way no human body could.

Hart says it was unclear if the app should be able to generate images of real people in line with Meta’s rules. Regardless if it was allowed to, it was doing so. Previously, the company said it wants “people to know when they see posts that have been made with AI” and that it “automatically adds labels to some user-generated content when AI is detected.” Despite this company statement, no obvious indication or label stating that the material was generated by AI.

Meta spokesperson Tracy Clayton wrote a brief statement in response to Hart’s questions about the purpose of the feature, if the company considers the stories fictional or factual, what safeguards are in place, and if the images of public figures comply with its own AI-content policies.

“We’re testing a daily feed that proactively shares tips, content, and recommendations tailored to your interests. The goal is to suggest what’s most relevant to you – such as fitness advice, meal plans, or other insights – before you even have to ask.”

Later, Clayton sent an updated statement removing the word “proactively.” A third statement was sent by Clayton later the same day: “This was a test for a limited number of users, and it will be deprecated. Meta has no plans to move forward with this feature.”

Sources:

The Times of India: After 8,000 layoffs, Meta tells 7,000 employees in memo: You have been identified as someone who can
The Verge: Meta made its own AI-generated clickbait news feed/Meta said it would pull the feature after The Verge asked questions about it

Featured Image Courtesy of Cory Doctorow’s Flickr Page – Creative Commons License