Workers Filed $64,000 in Pay Disputes at AI Training Startup Handshake | Invesloan.com

Joel Thomas was hooked the moment he saw the pay rate.

Handshake AI, the startup behind a career-services app that his university recommended, invited him to apply in August for AI training work paying $50 an hour or more. That would be enough, he figured, to cover his college expenses for a whole year.

Now, Thomas says, he is owed about $11,000 in back pay for hundreds of hours of work he did on Handshake’s platform. Handshake has denied his appeals. The company told him there was no record of his work on one project during the third week of August, though a Handshake pay statement showed he had completed hundreds of tasks that week.

He isn’t alone. Nearly two dozen current and former contractors told Business Insider that Handshake this year underpaid them by hundreds or thousands of dollars each. Disputes from those contractors over the past six months totaled $64,000, the contractors said.

Handshake said that more than 188,000 contractors have earned over $500 million for AI training projects on its platform. Startups like Handshake have recruited armies of students and young professionals around the world to train AI models for tech giants such as Meta, Google, OpenAI, and Anthropic. The booming AI-training industry has come under increasing scrutiny this year for the way it treats that sprawling contractor workforce.

Pay disputes at Handshake arose for several reasons, a Business Insider investigation found. In many cases, contractors’ pay didn’t match the dollar amounts they expected because of differences in the way Handshake and its clients tracked time and pay.

Former staffers said insufficient full-time staff and pressure to deliver training data at a breakneck speed also contributed to the payment problems. Project managers didn’t have time to address contractors’ questions — such as when some contractors found that instructions for a task were confusing or contradictory, former employees and contractors said. Once a contractor completed a task, it was subject to review; if a contractor didn’t understand the instructions and the task didn’t pass the review, the contractor wouldn’t be paid for the work, former employees said.

On some marquee projects, one person could be responsible for managing up to 1,000 contractors, former staffers familiar with the operations said. Business Insider reviewed bank statements, pay stubs, and hundreds of email and Slack messages between contractors and the company.

Thomas, a soft-spoken junior who in his downtime likes to hike near his northern Virginia home, said he has tried to stay positive.

“It’s hard to do because it’s still a lot of money,” he said.

‘Competitive pay, clear expectations’

Business Insider first reported in March that dozens of contractors were complaining that Handshake had denied them payments, and that one contractor had won a $6,475 judgment in small-claims court. Handshake, in many cases, told contractors in automated messages that their accounts had been closed because they had violated platform rules. In October, after Business Insider sent Handshake detailed questions for this story, the company introduced changes to its platform that it said would improve pay transparency for contractors, whom Handshake calls “fellows.”

Handshake said it has made contractor pay clearer with a new payments dashboard, more explanations for adjustments, and an easier dispute process. Handshake said it is also rolling out a new live timer to track work on third-party platforms. The company told Business Insider that any contractor who believes a payment decision was wrong, or who has new information, can raise it again with the company’s support team.

“Our Fellows are the heart of Handshake AI,” the company wrote in an October blog post. “As part of being a Fellow you should always know what you earned, how it was calculated and when it will arrive. That’s a basic expectation for anyone doing work with Handshake AI, and we’re committed to meeting it for every Fellow.”

Handshake declined to comment on Thomas’ situation, saying it doesn’t comment on individual cases.

“We take every dispute seriously,” the company spokesperson said.

On podcasts and in LinkedIn posts, Handshake CEO Garrett Lord has repeatedly pitched transparent, reliable pay as one of the company’s advantages over its rivals, which include Scale AI and Mercor. He has said complaints about late payments at other AI-training companies contributed to Handshake’s decision to enter the business.

“Most AI gigs = shifting rates, zero transparency,” Lord wrote in a LinkedIn post last year. “Handshake AI = competitive pay, clear expectations, and real dialogue. That’s what experts deserve.”

The AI pivot

Thomas was a sophomore at Liberty University in Lynchburg, Virginia, when the school directed him to Handshake’s platform, which hosts a virtual career fair for the school. He signed up for an account on Handshake in January.

On its website, Liberty University says it doesn’t endorse content, jobs, or employers on Handshake.

“Liberty University cannot guarantee that use of any of these products alone will guarantee the user a positive outcome,” it says. The school didn’t respond to requests for comment.

Not long after he signed up, Thomas began receiving invitations from Handshake to work on AI training projects, which the company pivoted into last year.

Handshake, founded in 2014 by three Michigan Tech students, had started out as a career platform. Students who linked their Handshake accounts to their school could search and apply for internships and entry-level jobs, register for career fairs, and connect with employers and alumni, among other services.

Colleges and universities paid Handshake a fee to connect their students with prospective employers. As the AI boom became a magnet for investor cash, Handshake saw a bigger opportunity.

In January 2025, Handshake expanded into AI training, tapping its network of students and young alumni to help frontier labs train and evaluate their AI models. Handshake also rebranded, adding AI to its name.

A slew of startups like Handshake pay gig workers to help make large language AI models like ChatGPT smarter. These contractors do tasks like writing expert answers to sample prompts, designing edge-case questions to test models and expose weaknesses, and rating AI-generated responses.

Handshake’s AI business has exploded: Lord said in August that its AI unit had grown to more than $1 billion in gross annualized revenue. The majority of Handshake’s revenue, former staffers told Business Insider, now comes from AI-training contracts with five clients: Meta, Google, OpenAI, Nvidia, and Anthropic.

Racking up the hours

“Great news!” the email said. “We have a new project through Handshake AI that we think you’d be interested in.”

Thomas had taken on a couple of small Handshake AI training projects in the spring, but stopped because the hourly rates were low — around $17 per hour or less. This invitation and another that arrived in August were more enticing. These two projects would pay $50 per hour and $65 per task, respectively.

He got right to work, planning to log as many hours as possible before his junior year started.

Thomas recalls that he would wake around 8 a.m. or 9 a.m., have a coffee, and begin working at his laptop, sitting at the white desk in his bedroom. Wearing AirPods, he cycled through R&B from Chris Brown and Bryson Tiller. Back at his desk after gym and dinner, he worked until midnight or 1 a.m.

Inside Handshake, staffers were pulling similarly long hours, former employees said. The company was scrambling to deliver an ever-growing amount of training data to its clients.

Project managers used AI coding tools like Claude or Lovable to generate instructions and prompts for contractors, but they were sometimes contradictory, leading contractors to seek clarification on their tasks, former employees said.

Questions from contractors piled up on Slack, and project managers didn’t have time to answer them all, the former employees said.

One of Thomas’ projects was called Project Folio, a study of ChatGPT usage patterns for which he was asked to upload his past chat logs with ChatGPT. For another project, he had to evaluate ChatGPT responses.

As he racked up hours, Thomas could see his earnings ticking up on his Handshake dashboard. At one point, the number topped $19,000, he says.

There was something important about that dashboard that Thomas says he didn’t know.

Like most AI-training platforms, Handshake pays contractors in units that sound intuitive — dollars per hour, or dollars per task.

The dollar figures that Thomas and other Handshake contractors saw on their dashboards were estimates based on the time they had manually logged by starting and stopping a timer. The AI company paying for the project had another timer logging their work — a timer that the contractors couldn’t see, former Handshake employees told Business Insider.

Before paying contractors, Handshake compared the contractor’s self-reported hours with the client’s timer.

Handshake didn’t tell contractors what counted as billable time, or how it was calculated by the client, former contractors and employees said.

“Final pay reflects the work our systems verify,” the Handshake spokesperson told Business Insider. “Where logged time doesn’t match the work, we review it, and where we find inflated time we don’t pay for it.”

Handshake said the company is committed to paying contractors accurately for the time they work and making sure it’s clear how much they will be paid.

When a contractor disagrees with a payment decision, “a dedicated team independent of the project reviews it in full and can restore all or part of the amount,” the Handshake spokesperson said.

‘This is ridiculous’

Thomas transferred to the University of Mary Washington in the fall. He was in the library during his first week of classes there when he saw a payment come through from Handshake. It was far lower than he expected.

Angry and confused, he went to Reddit, where AI trainers often talk about their work for Handshake and its competitors. He posted a screenshot of the payment and wrote: “This is such a scum thing to do and I am unsure what I should do. Just want to expose them as much as possible because this is ridiculous.”

Another user suggested someone at the company whom Thomas could contact. Thomas says the response he received from Handshake wasn’t reassuring.

By the time his next payment arrived, Thomas was braced for another disappointment. It, too, came up short, Thomas said. In all, he was paid $8,000 for what he believed was $19,000 worth of work.

He hadn’t captured a screenshot when his dashboard showed $19,000, but he had captured one when it showed $15,741.23 “awaiting payout” on his account. He emailed the screenshot to Handshake support staff.

“I highly suggest you check your systems,” Thomas wrote in an August 27 email.

Handshake told him that the “awaiting payout” figure was an estimate based on the times he had clocked in and out on the Handshake timer, and that before final payment, the company had checked his time against his completed tasks. For the period of August 17 to August 23, the company told him, “our project leads did not find any tasking that took place during this time frame.”

This explanation seemed to conflict with a pay statement Thomas received for that period, showing that he completed 428 tasks for Project Novel and would receive no pay for them.

A Handshake spokesperson told Business Insider that contractors aren’t paid for tasks if the tasks aren’t approved by reviewers, and that the phrase “no tasking took place” is shorthand for “no verified work.” The company may deny payment for time that exceeds a project cap, or for logged time that doesn’t match the recorded work, the spokesperson added.

Handshake notified Thomas that it had deactivated his account because of irregular activity “that does not align with our platform policies and usage expectations.”

Thomas remains cautiously optimistic about how AI might shape his career. He said he sees AI as a useful tool for assisting work, particularly in science, technology, engineering, and math. At the same time, he worries that AI will replace too many jobs and could make it harder for some people to find work.

He’s still open to helping train AI models. But he’s given up on getting any more pay from Handshake.