Big Tech doesn’t want you to know about these workers behind AI…
Recently, hundreds of Ireland-based workers critical to fine-tuning Meta’s AI products were laid off. No warning. No severance for most. And to add insult to injury, they would not be able to pursue another role at Meta due to a six-month “cooldown period” policy.
How did Meta manage to leave them in the dust so easily? That’s because these content moderators and data annotators actually work for Covalen, a third-party contracting company that seems to work primarily on Meta projects.
This is a common practice within the AI supply chain: a labyrinth of hundreds of companies and thousands of data workers and content moderators all around the world are collecting and annotating data, training AI models, and stepping in when AI tools fall short of a company’s promises.
Big Tech companies don’t want you to know that they are the real intelligence behind artificial intelligence. They definitely don’t want you to know how these workers are paid and treated.
That’s why we developed the Data Work Landscape, which maps 116 companies; 1,900+ investor, client, and partner relationships; and the hidden supply chain powering AI. But to truly understand the impact of this layered system, we gathered five stories from workers at these companies, drawn from our own work and public reporting.
Right here in the US they pay as low as possible. Telus pays between $12 and $14 an hour. I’m paid $14 in Missouri, and I expect when the minimum wage goes up (to $15) at the end of the year that I get let go from that because they decide we’re not going to pay anyone that much.
Kirn Gill II Search Quality Rater working on Google products at Telus
Kirn expects Telus to lay him off soon
Kirn Gill II is a US-based search quality rater who is preparing for his own layoff. Working through Telus, Kirn helps Google improve search algorithms by comparing two versions of the same search and ranking to determine which version better meets the user’s needs.
Despite doing vital work for one of the largest companies in the world, Kirn is paid $14 per hour—about $ 7 per hour below Missouri’s living wage. Other Telus employees across the U.S. say that it’s difficult to live on these low wages. Some don’t even receive the full wages for their work when it takes them longer to complete the task than Telus estimated.
When the minimum wage was due to increase to $15 per hour, researchers from TechEquity and the Communication Workers of America spoke with Kirn, but Kirn says that probably just means he and his co-workers will be let go. These companies would rather cut workers than pay them more.
That’s because, as Kirn conveyed, these companies don’t see them as human. “If there’s anything I wanted the general public to know,” he told researchers, “it is that there are low-paid people who are not even treated as humans—just little more than employee ID numbers—out there making the 1 billion dollar, trillion dollar AI systems…”
Lilian is penalized for Appen’s dysfunction… again and again
Located in Brazil, Lilian also faces dehumanization and instability. If all goes well, she earns about $250 per month doing data work through Appen, according to her interview with The Intercept Brasil. Appen is an Australia-based company that contracts for Salesforce, Amazon, and the US Department of Defense.
The job promised flexible hours—at a cost. Lillian works on demand, picking up tasks as they become available. Lillian does not get paid for the time she spends waiting for tasks. And when tasks do appear, she has to be ready to jump on or risk losing them to another worker.
At one point, Lilian received a low score on her tasks—60% compared to her usual 100%—and was automatically kicked off the platform. She desperately tried to reach someone at Appen but was only able to log back in with help from other data workers. When she finally did, she found the platform counted two hours of work as six minutes. She complained. She got more work. She received a paycheck that was still $30 short.
This isn’t unusual for Appen. The Guardian reported in 2023 that when Appen was migrating to its new worker platform CrowdGen, as many as a third of all data workers remained unpaid, with workers instructed to continue working on projects with no clarity on when payments would be made.
Eliza is vulnerable and getting traumatized at Teleperformance
In Portugal, Eliza’s work is tied directly to her physical well-being. The company she works for, Teleperformance, provides housing for its employees—in exchange for lower wages, according to Eliza. But what happens if workers are laid off or fired?
It’s a precarious situation for someone doing some of the hardest work in tech. Eliza is a content moderator who spends 8 hours per day, 5 days per week, reading hate speech and reviewing graphic images to determine whether a post violates a platform’s policies. The identification work done by content moderators is often used to train AI models as well.
It’s arduous, traumatizing work, but Eliza only gets a 30-minute “wellness” break every day, a common practice at Teleperformance. In 2025, Ghana-based workers at a company acquired by Teleperformance reported inadequate mental health care support to deal with the psychological trauma from content moderation work.
That, combined with the low pay, is why Eliza and her coworkers went on strike in 2023.
Despite Teleperformance signing an agreement with UNI Global Union to secure workers’ right to collective bargaining, the local president of Teleperformance thinks they make good enough money, as Eliza told TechEquity. So he wasn’t interested in raising wages. Nevertheless, Eliza remains hopeful for positive change in the future as workers continue to band together and people become more aware of the issues in the AI supply chain.
Junbee makes below minimum wage working for Remotasks
The informal settlement of Cagayan de Oro, Philippines, has become a hotbed of low-wage data work.
There, Junbee works for Remotasks, a subsidiary of Scale AI. His first few weeks working for Remotasks were spent at their local headquarters, where he was crammed into a windowless room, training all day under the watchful eyes of supervisors, learning how to annotate data. This involves decrypting images, videos, and even audio for AI so that the system can “learn” what, for instance, avocado toast looks like.
As the name Remotasks would imply, Junbee then went “remote.” But without his own computer, he mostly works out of internet cafes.
Junbee told Equal Times that he works between eight and ten hours per day doing tasks for companies like Apple, DoorDash, and even the Government of Qatar. Despite an impressive client list, Junbee earns only about $7 per day, which is less than the legal minimum.
Other Remotask workers in the Philippines have also reported below-minimum wages as well as delayed and withheld payments. However, there’s little recourse for workers at foreign companies like Remotasks, and that’s by design. For instance, the U.S. Department of Labor investigated Scale AI’s compliance with the Fair Labor Standards Act during the Biden Administration, but the investigation was reportedly closed under the Trump Administration.
Despite these issues, Junbee and others continue working for Remotasks as they have few other legal opportunities to make a living.
Sonia is guilted by Sama into accepting terrible working conditions
The mother of two young children, Sonia also saw an opportunity to provide for her family through the AI supply chain. The reality of working for the Meta subcontractor Sama, though, was brutal—despite the company advertising itself as an “ethical data pipeline” provider.
For two years, she spent sometimes ten hours per day seeing the worst content on the internet: child abuse, human mutilation, racist attacks, and more.
In exchange, she was paid a measly $2/hour and was barely allowed to take breaks. When she or other workers complained about conditions, Sama management would pressure them, saying that they needed to work hard or else they’d lose the contract with Meta and everyone would lose their jobs. That threat would prove prophetic. Years later, other Sama contract workers were let go when they went public about privacy issues with Meta’s smart glasses.
Despite these actions, Sonia didn’t back down and has since become an advocate and researcher in the hopes of bettering working conditions for other content moderators and data workers. And she is not alone.
The story of data workers isn’t over
From the US to the Philippines, Kenya to Ireland, the workers behind AI are tired of toiling while Big Tech makes billions.
One of the biggest challenges in recognizing and protecting workers’ rights in the AI supply chain is untangling the web of contractors and subcontractors. This is not an accident; it’s by design. This layered system is meant to enable companies to extract from workers and protect Big Tech and other prominent companies from responsibility.
That’s why we recently updated the Data Work Landscape to include detailed profiles of 10 of the industry’s most prominent players. In our accompanying blog post, we detail the connections and trends we unearthed while updating the landscape.
With this, we hope that journalists and researchers can continue making important connections, advocates and governments can be more equipped to hold bad actors responsible, workers can find who they’re actually working for, and that everyday people can grasp a fraction of what’s behind the AI tools they use and that are used on them. We’re all in this together. That’s how people changed the garment industry, and that’s how we can make change in the AI supply chain.
This project is just a snapshot of this growing industry; we need your help to complete the picture. If you’re familiar with or work in the AI data work industry, please share your experiences with us using this form.