Handshake AI Faces a Misclassification Suit Over Its Annotators

A wage-and-hour suit filed in federal court in San Francisco in late July is asking a question the AI training industry has so far avoided answering in a courtroom: are the people who annotate data for frontier models independent contractors, or are they employees who have been paid as if they were not? The case is Boggs v. Stryder Corp. d/b/a Handshake et al., Case No. 3:26-cv-07498, filed July 20, 2026, in the U.S. District Court for the Northern District of California. Summons were returned executed on August 5, and the defendants are not yet due to answer. Nothing in the complaint has been tested, and Handshake has not filed its response.

The lead plaintiff is Sean Boggs, represented by attorney Trenton Ross Kashima. The named defendants are Stryder Corp., which does business as Handshake, and Handshake AI Solutions, LLC. The primary cause of action is brought under 29 U.S.C. § 206, the minimum wage provision of the Fair Labor Standards Act, alongside claims under the California Labor Code. Boggs seeks to recover unpaid wages on behalf of himself and a proposed class of similarly situated annotators. The core legal theory is misclassification: that Handshake treated the workers as independent contractors while exercising the degree of control that, under both federal and California law, defines an employment relationship.

The complaint describes a working arrangement that, as alleged, leaves annotators very little of the independence the contractor label implies. According to the filing, Handshake sets pay rates unilaterally rather than negotiating them, requires workers to perform their tasks on the company's own proprietary platform rather than tools of their choosing, and imposes strict caps on how many hours a worker may bill in a given week. The complaint further alleges that workers were required to attend onboarding sessions and meetings for which they were not compensated, and that contractors who raised concerns about pay faced retaliation. Each of these is an allegation; Handshake has not responded to them in court, and the company has not publicly addressed the specific claims.

Wage claims brought under 29 U.S.C. § 206 and the California Labor Code follow a fairly standard remedy structure, and the Boggs complaint's stated cause of action is consistent with it. Beyond recovering unpaid wages for hours worked, including time the complaint characterizes as uncompensated onboarding and meetings, plaintiffs bringing this type of claim typically pursue liquidated damages, which can effectively double the wage recovery under the FLSA, along with attorney's fees and injunctive relief that would bar the challenged classification practice going forward. The specific relief requested in this filing was not detailed in the materials reviewed for this article, and readers should treat that structure as the general shape of this kind of suit rather than a confirmed list of demands.

California's ABC test raises the bar

Venue matters a great deal here. California applies the ABC test to classification questions, a standard that presumes a worker is an employee and puts the burden on the hiring entity to prove otherwise. To defend a contractor classification, a company must satisfy all three prongs, not merely most of them. Prong A asks whether the worker is free from the control and direction of the hiring entity in performing the work. Prong B asks whether the work falls outside the usual course of the hiring entity's business. Prong C asks whether the worker is customarily engaged in an independently established trade or business of the same nature. Read against the allegations, each prong presents a distinct problem: mandated tooling, dictated rates, and weekly hour caps go directly to prong A; if Handshake AI Solutions exists to supply annotation services, annotation is arguably its usual course of business, which puts prong B in play; and prong C asks whether these workers run annotation businesses of their own, which is a difficult characterization for someone whose only client sets their rate and their schedule. A defendant that loses any one prong loses the classification.

The California claims are only part of the filing. The complaint's underlying cause of action rests on the FLSA, and federal courts use a different framework than the ABC test to decide whether a worker is an employee. Known as the economic realities test, it weighs the whole relationship rather than requiring a defendant to clear three separate prongs: whether the worker has a genuine opportunity for profit or loss based on their own initiative, whether they invest in their own equipment or tools, whether the work is permanent or project-based, and how much control the hiring entity holds over how and when the work gets done. That last factor is where the complaint's specific allegations are aimed: mandated tooling, unilateral rate setting, and capped weekly hours. The theory is that a worker who cannot set their own rate, choose their own platform, or work as many hours as they want is economically dependent on a single company in a way the FLSA treats as employment, regardless of the label on the contract.

Handshake spent most of its history as a campus recruiting network, the software many U.S. universities use to connect students with internships and entry-level roles. It reached a reported $3.5 billion valuation in early 2022 on that business. More recently the company moved substantial resources into AI data labeling and model training, marketing expert-level annotation tasks to the students and recent graduates already on its platform. In a previous podcast appearance, chief executive Garrett Lord described contractors on the platform earning between $100 and $125 per hour for specialized AI training work. The work itself is a step removed from the image most people have of data labeling: the fellowships Handshake advertises are not simple tagging or classification tasks, but are described as reviewing and grading model outputs for accuracy and reasoning, a category of work commonly called reinforcement learning from human feedback, and the company's own listings describe tasks that include writing and reviewing code, working through advanced mathematics, evaluating creative writing, and checking legal or medical content for accuracy. That pivot is what gives the case its shape. A recruiting network has a built-in supply of educated, credential-bearing workers who already trust the platform, and expert annotation is precisely the kind of work frontier labs are paying premium rates for. The suit tests whether the labor model attached to that supply holds up.

The Boggs complaint is not the first pay dispute involving Handshake's AI contractors, though the earlier matters are documented unevenly and should be read with that in mind. Contractors working on OpenAI-related projects reported in March 2026 that they were denied payment for completed work, in some cases after more than 50 hours on a project, on the grounds that they had violated platform rules. Workers described the withholding as permanent and said no meaningful appeal was available to them. Separately, a contractor who was removed from the platform for using AI tools to complete tasks, a practice the contractor said Handshake's own policy permitted, sued for unpaid wages and was reportedly awarded $6,475. Contractors have also described what they call shadow capping: completing assigned work, logging the hours, and then receiving less than the logged time implied because of adjustments made on the platform side without notice. These accounts come from contractors rather than from court records or company statements, and they have not been adjudicated.

The reason this case reaches beyond one company is that the contractor model is load-bearing for the entire AI training supply chain. Frontier models require enormous volumes of human annotation, evaluation, and red-teaming, and nearly every platform that supplies that labor does so through contractors rather than employees. Reclassification changes the arithmetic: employees carry minimum wage floors, overtime obligations, payroll taxes, workers' compensation, and in California a set of expense-reimbursement and meal-and-rest requirements that have generated substantial liability in prior gig-economy litigation, and applied across an annotation workforce, those costs are not marginal. A plaintiff win in the Northern District would not automatically reclassify anyone outside the class. It would, however, give other plaintiffs' firms a template and a venue, and it would put pressure on every platform whose operating model depends on directing contractor work closely while paying for it loosely. The industry has watched this pattern before with ride-hailing and delivery; this is the first time it has arrived squarely at AI training.

What contractors should document now

Whatever happens to this particular case, the practical lesson for anyone doing annotation work is the same, and it costs nothing to act on.

Keep your own record of hours. Do not rely solely on the platform's timer or dashboard, which can be adjusted after the fact and which you generally cannot export once an account is closed. A simple log with dates, project names, and start and stop times is enough, and it is the single most useful thing to have if a payment is ever disputed.

Track unpaid time separately. Onboarding, mandatory training, calibration sessions, and required meetings are hours worked even when a platform does not treat them as billable. If classification is ever litigated, that time is part of the claim.

Save the terms you agreed to. Screenshot the rate card, the policy pages, and any project brief that states what is permitted. Policies change, and the version in force when you did the work is the one that matters. The contractor in the AI-tools dispute above prevailed on exactly that point.

Reconcile every payment against your own log. Where the amount differs from what you recorded, ask for a written explanation and keep the reply. A pattern of unexplained adjustments is far more persuasive than a single one.

Finally, spread your work across more than one platform. A classification fight, an account suspension, or a payment freeze at any single client should not be able to stop your income. That has been sound practice in this industry for a while, and cases like this one are a reminder why.

Sources

1. Boggs v. Stryder Corp. d/b/a Handshake et al., Case No. 3:26-cv-07498, U.S. District Court, Northern District of California, filed July 20, 2026; summons returned executed August 5, 2026
2. 29 U.S.C. § 206 (Fair Labor Standards Act, minimum wage) and California Labor Code
3. California ABC test as codified at Cal. Lab. Code § 2775 (AB 5)
4. FLSA "economic realities" test, applied by federal courts under 29 U.S.C. § 206 et seq.
5. Contractor accounts of withheld payment on OpenAI-related projects, March 2026, and of undisclosed hour adjustments, not independently verified

Related reading

Handshake AI review: how the fellowship program works, who qualifies, and what the roles pay.

Taxes for annotation work: what contractor status means for your filing, and what changes if it does not hold.

Is AI training legit?: how to check a platform's payment and appeals record before you start working.

Running multiple platforms at once: the portfolio approach that keeps one platform's dispute from stopping your income.

Best AI training platforms compared: how the major platforms handle pay, appeals, and account decisions.