How AI’s Labor Supply Chains Fail Workers

In April, Kenyan data worker Grace* worked the night shift for Sama — a San Francisco-based outsourcing firm with physical offices in Nairobi that supplied trainers for Meta’s AI systems, including Meta’s AI-powered Ray‑Ban glasses.

“You'd get to the office at 10 p.m., do a biometric for your check-in, and put your Zoho on,” she told Tech Policy Press.

According to Grace, Zoho is the software Sama uses to monitor workers’ attendance. Alongside this ran Meta’s own tracking system — which records everything from bathroom breaks to daily productivity levels — until she logged off at 6 a.m.

On April 16, that routine came to an end. Sama announced it was making Grace and more than 1,000 of her colleagues redundant after receiving a notice from Meta to end its engagement in Nairobi. Grace and several coworkers say Meta had pulled a major contract citing only compliance with Kenyan employment law.

“From what they told us, they said it was because of the exposé from the Swedish journalists,” she said. Back in February, Swedish publication Svenska Dagbladet published an exposé about how Meta’s glasses captured intimate audio-visual data with little transparency about how it is stored, used, or monetized.

“After the exposé, they started looking for people who could have leaked the information,” said Grace. “They were like, if we find you, we are going to deal with you.”

“That very night,” she recalls. “The day the exposé came out, around midnight, Meta had cleared all the queues – nothing was coming to our end.”

This precarity is not incidental; it is structural.

The outsourcing maze

The data annotation and labeling market is one of the fastest-growing in the world, with intelligence firm Global Insight Services predicting market expansion from $1.2 billion in 2024 to $10.2 billion by 2034.

Despite those projected profits, worker precarity is a defining characteristic of the industry — embedded in the way labor is organized and supplied to Big Tech through multiple layers of outsourcing.

A March report by SOMO, a Dutch nonprofit research center that investigates the social and environmental impact of multinational corporations, detailed Big Tech’s impact on worker conditions worldwide and found that “almost 500 companies are currently active in AI data collection and labeling,” and that “Amazon, Google, Meta, Microsoft, and Nvidia combined use at least 30 intermediary companies for data work.”

The intermediaries operate as different types of businesses. Sama, for example, is a business process outsourcing firm that employs workers to deliver services to Meta. Others — like Outlier, owned by Scale AI — operate as micro-tasking or crowdworking platforms where workers are hired as independent contractors and are paid on a per-task basis.

The outsourcing maze extends even further in other AI labor supply chains: Contracts are passed through subsidiaries in India or the Philippines, or shell companies registered in tax havens. For workers, this means chasing answers across borders, through firms that may not even have a physical office.

“If I was working directly for Meta, I can tell you for free, these are not the challenges I’m going to speak about,” said Joan Kinyua, president of the Data Labelers’ Association. “I'm going to be paid well. I am going to be protected against the harms that these things cause.”

That gap between direct employment and outsourced work is widening as Big Tech begins to fold human labor into the cloud itself. As the SOMO report notes, companies like Amazon are “making data work an on‑demand component of AI development.”

Amazon’s Augmented AI shows how this works: when a business buys cloud services from AWS, it doesn’t just get servers and software. It can also hire workers through recommended vendors such as Cogito or iMerit. The catch, as SOMO points out, is that platforms “do not appear to provide information about or screen suppliers for workers’ conditions and rights.”

Grace confirms that Big Tech was more than just a customer — Sama employees used Meta’s tools and were integrated into the company's workflows when working on Meta’s AI products. “Everything was on their software, everything was on their apps. There was only time tracking from our end,” she said.

In this way, access to labor is streamlined but clients never have to think about who is actually doing the work. Annotators, labelers, and trainers — most either from or based in the Global Majority countries — are largely invisible despite forming the foundation of the AI industry.

Accountability across the chain

In Nairobi, the work may have stopped, but the question of who is responsible for the workers themselves was never fully answered. Sama’s April 16 press release announcing the mass layoffs spoke of compliance with Section 40 of Kenya’s Employment Act and mentioned that the firm was “actively supporting affected employees with care and respect.”

“The only thing that they offered us was wellness,” said Grace. “They said counselors are available.”

Sama employees were offered a wellness benefit, which included access to trained counselors on-site but Grace said that every benefit came with strings. “These [counselors] are the same people that were hanging out with the top management during lunch,” she explains.

In the AI labor economy, responsibility is so diluted that it diffuses across companies and regulators. Meta said Sama is the employer. Sama points back to compliance with Kenyan law. Kenyan officials propagate a national growth narrative that treats outsourcing as an opportunity — seen clearly in the government's celebration of a $10 million pledge from ADEC Innovations, a global outsourcing provider headquartered in the Philippines, tied to 2,000 new business process outsourcing jobs.

“In countries where they're very serious about their labor laws, Big Tech follows those laws,” said Joan. “But when it comes to a country like Kenya, there's nothing to follow.”

Kings Korodi, co-founder of Techworker Community Africa, a Nairobi-based advocacy group that organizes data and platform workers across the continent, echoes those sentiments: “The government should be responsible for regulatory oversight and labor protection, but they lag behind because they're incentivized to attract outsourcing investment.”

But researchers aren’t letting Big Tech off the hook. Margarida Silva, author of the SOMO report, said Big Tech cannot avoid taking responsibility for labor conditions because “they are creating the need, and they're creating the conditions under which this work is established,” she said. “They could, if they wanted to, use that power to ensure that workers have good working conditions, but that doesn’t seem to be happening.”

Measured into obsolescence

Grace doesn’t know what to do next. She had worked for Sama for over a decade and is now faced with starting from scratch. “I should be very far by now but I am not. I am still stagnant.”

Even before the layoffs, her salary — 30,000 Kenyan shillings (about $230) per month before tax — was below subsistence levels. “It wasn’t really enough but it was something, like, at least you’re not totally broke,” she said. 

According to Kings, the problem goes deeper than wages. He points to the way workers are governed by algorithmic management, including “productivity metrics, accuracy scores, and automated task allocation pipelines that directly shape their pace of work, income stability, and continued employment.”

What both Grace and Kings are pointing to is a system that reduces workers to interchangeable units — monitored, measured, and laid off when contracts shift, or corporate priorities change.

This is what disposability in the AI labor economy looks like in practice. Contracts can be pulled at a moment’s notice with no avenue for recourse, leaving workers adrift. These are workers who are essential to AI’s development but are completely stripped of their power in the process.

For Grace, that loss of power is visceral. The sudden layoff has forced her to confront the reality that she is back at the starting blocks, searching for direction after the industry she helped sustain erased her overnight.

“Right now, I'm just finding my landing space. I'm starting to think about what my next steps are,” she said. “I need to find something that's motivating; something I'll be excited and happy to do.”

Despite fragmented accountability, it is clear who bears the consequences of decisions in this supply chain. When responsibility is layered between Big Tech, BPO firms, and the government, it does not land on those with power — it lands on the workers.

Meta and Sama were contacted for comment, but no response was received by the time of publication.

*Names changed to protect sources