Ricardo Levario spent almost a decade working in elementary schools, but by 2023, he’d finally had enough. He quit his job without a backup option, and he was a month into the job search when he received a cryptic message from a Linkedin recruiter looking for teachers. Levario started at GlobalLogic on May 1, one of the first 25 “super raters” assigned to work on Google’s AI products. Business was booming. By summer, GlobalLogic was hiring up to 20 people a week. When that wasn’t enough, they hired staffing agencies to bring on even more.

At first, Levario was grateful. He had enjoyed playing around with ChatGPT when it became publicly available, and he was excited about the prospect of a new career in tech. His background in education felt relevant to the day-to-day work—he often found himself grading and rewriting AI responses, or drafting detailed instructions used to train Google’s products.

But the problems were becoming harder to ignore. Classified as temporary for an entire year, Levario was denied paid time off and holiday pay. He was eventually offered a small raise to go with his promotion, but 70 percent of other promoted workers were not. Subcontractors, hired through staffing agencies, made only around two-thirds of their coworkers’ wages after the agencies took their cut. And in the first months of 2025, hundreds of workers were laid off.

GlobalLogic workers might be developing Google products, but they don’t enjoy Google perks. “There’s no ping-pong tables and free lunch for these people,” says Shannon Wait, an organizer with the Alphabet Workers Union. GlobalLogic, a subsidiary of the multinational conglomerate Hitachi, employs about 1,400 raters under a contract with Google, which is a subsidiary of Alphabet. It’s a common setup—since 2018, contractors have made up the majority of the workforce at Alphabet, working under inferior conditions and the constant threat of losing their jobs. That tangled corporate web can deter solidarity on the shop floor, but GlobalLogic workers have moved to organize anyway.

Levario was one of the most prominent of those organizers—until this February, when he was fired amid mounting conflicts over the union campaign. While he and others have filed complaints with the NLRB, they have little legal recourse given Trump’s incapacitation of the agency. Still, they haven’t given up: In April, GlobalLogic employees released a demand letter calling for better pay parity and job security. “Unions brought about the NLRB, not the other way around,” says GlobalLogic subcontractor Rachael Sawyer. “Our power was never in the law.”

The number of Google contractors working on various projects isn’t public knowledge, but the company may rely on as many as 12,000 AI workers across seven to 10 different contractors, Wait, who started working with GlobalLogic raters after they reached out to her last February, estimates. Other artificial intelligence engines depend on thousands more. At GlobalLogic, which declined to comment for this story, workers are assigned to support Google engineers working on projects that include Gemini, Google’s AI “assistant.”

Current Issue

December 2025 Issue

In some ways, this type of work is not new—engines like Google have long relied on underpaid raters to train their search algorithms—but the AI boom has led to exponential unregulated growth in this workforce, as tech juggernauts pour billions of dollars into the race to capture market share. In 2023, a Time investigation found that OpenAI had paid Kenyan contractors less than $2 an hour to watch and identify violent and abusive content. The workers, who also rated content for Meta, lost their jobs when they tried to organize. A lawsuit against Meta is ongoing. On April 30, moderators launched a global trade union alliance across nine countries; so far, the United States is not one of them.

GlobalLogic workers, who are based in the United States, do make more money than their Kenyan peers, with wages for generalist raters starting at $16 an hour. So-called “super raters” are paid $20 or more an hour to do the same kind of work, because they usually have master’s or doctoral degrees—although that’s well under the average for American workers with comparable qualifications. In other ways, though, the work can be much the same across national borders. Rachael Sawyer works on a project training Google’s AI to filter out hateful and violent content, including child sexual abuse material. At one point, she received a waiver acknowledging that the work can be dangerous for mental health—it’s been known to cause PTSD. She decided not to sign the waiver, and GlobalLogic seems to have let the matter go, but conditions remain the same.

“When I tell people about my job, they think, Oh, I thought robots did that,” Sawyer says. “No, we’re human. We’re treated like robots, but we’re human.”

Sawyer has worked as a tech contractor since 2020. As she moved from one “unstable, underpaid, invisible job” to the next, she says, what got her through the day was being able to reach out to her coworkers. That was especially true at GlobalLogic, where she started last spring. The company had long encouraged remote employees to connect with each other online. Then, on January 21, Levario and other union members used the company’s social channels to distribute a survey on compensation and working conditions to all 1,800 (the number then) of their coworkers. Two weeks later, GlobalLogic announced a new policy: The use of social channels during working hours would be forbidden.

GlobalLogic’s broad restrictions on socializing, coming so soon after the survey was distributed, effectively quelched most organizing conversations. In some cases, restrictions on talking at work may be legal. But if workers are allowed to talk about any non-work topics on the job, the National Labor Relations Act says that they have to be allowed to talk about working conditions, too, and rules on socialization that “chill” employees’ right to organize are presumptively unlawful.

During an all-hands meeting, a GlobalLogic manager encouraged raters to continue to chat about movies, books, food, and pets. “Talk around what’s fair—money and days off and benefits and stuff like that—that’s not really something that is really work-appropriate to share with everyone,” she said in a recording reviewed by The Nation. “We can’t bring in negativity, things that could stir the pot and upset people.”

The next day, Ricardo Levario filed a whistleblower complaint with Hitachi, GlobalLogic’s parent company. At 8:59 Monday morning, the head of GlobalLogic HR responded to Levario’s complaint—and by 1 pm, he had lost his job.

In the aftermath, union membership tripled. When super rater Michael Bailey’s team was told that their workloads would be increasing, Bailey commented that it would be nice if his laid-off colleagues were still around to help. “No one is restricting your right to speak about the conditions,” a leader told him, in an exchange obtained by The Nation, but added, “It’s a time and place. You wouldn’t go to church and talk about drinking 40 oz. malt liquor and a dance contest during altar call.” Bailey was eventually summoned to a disciplinary meeting, which he recorded with consent. His efforts to “rally the troops,” he was told, were having a negative effect. “You have to make the choice on whether to comply or to keep doing what you’re doing,” the leader told him. “If you keep doing what you’re doing, then we can’t be surprised with how things play out afterwards.”

He shared the audio in a private chat for union members only. When a leader snuck into the chat and saw the recording, Bailey was fired too.

In the weeks since, restrictions on opportunities for organizing have only intensified. Content moderators are now forbidden from asking each other too many questions, even as they’re expected to maintain a 97 percent accuracy rate. GlobalLogic plans to require low-level leaders to oversee performance reviews, and when a worker warned her colleagues that taking on more management duties could affect their ability to join the union, she was fired.

Both Ricardo Levario and Michael Bailey have filed complaints with the NLRB, and the third worker plans to do so as well. But since Trump fired NLRB member Gwynne Wilcox a week after he took office, the board, lacking a quorum, has been unable to function. In the months since, Wilcox has been reinstated twice and removed twice, most recently by the Supreme Court, which issued a stay keeping Wilcox out of office until they review the case. Even in a best-case scenario, two empty NLRB spots await Trump appointees; at worst, the Supreme Court dismantles the board altogether.

“These individuals are employees of Hitachi, not Alphabet,” Google said in a statement to The Nation. “As the employer, Hitachi is responsible for the employment conditions of their workers.” A recent NLRB decision cast doubt on that claim: This January, the board ruled that Google was the joint employer of a different group of contractors, which would have forced it to bargain with them—and maybe GlobalLogic workers, too. But in April, a court nullified the NLRB decision.

That means that if raters decide to fight for a contract, workers at each of the seven to 10 Google AI contractors will likely have to bargain separately. And if GlobalLogic workers manage to win, there’s nothing stopping Google from canceling its contract with the company. The labor model that Google uses, Shannon Wait, the AWU organizer, says, “is a union-busting tool in and of itself.”

As mass layoffs accumulate, tech companies have enjoyed ever-greater power to crack down on worker actions—internal criticism, petitions, talking to journalists—that they may once have tolerated. Meanwhile, their contractors have never enjoyed anything resembling job security.

When Rachael Sawyer first started the job at GlobalLogic, she was struck by the “culture of fear.” People would discover that they were laid off only when their logins didn’t work on Monday morning. The natural response, Sawyer says, was to freeze, to accept conditions as they were. “But when you come together and you start to organize, that freeze flips into a fight response,” she says. “It’s no wonder they don’t want us talking to each other.”