Robotaxis Are Scaling Fast. Their Test Drivers Are Paying the Price.

One former Zoox contractor described a test run through a San Francisco intersection in which the autonomous software braked hard for a pedestrian who had not stepped off the curb. The jolt threw her forward against the seatbelt. She went home with a strained neck and stayed out of the cab for six weeks. The account is one of five given to TechCrunch, all from current and former Zoox contractors who spoke on the condition of anonymity, and none of it appears in any OSHA filing.

The visible record, drawn from the Occupational Safety and Health Administration's Injury Tracking Application, counts more than two dozen injuries to Waymo and Zoox test drivers across 2024 and 2025: sprains, strains, whiplash, and at least a few cases where workers were sidelined for weeks or longer. The reported cause, in most of them, is the same. The autonomous software braked hard or swerved suddenly, often with a trained safety operator behind the wheel.

Transdev, which employs and manages Waymo's test drivers, logged 16 injuries tied to Waymo depot locations in three cities. Zoox, which runs its own fleet of modified Toyota Highlander SUVs through San Francisco and other test markets, reported as many as 8 worker injuries linked to hard braking in the same window.

Waymo and Zoox are not the only autonomous-vehicle operators in the country. They are two of the very few that show up in this dataset at all.

OSHA only requires establishments with 100 or more employees in higher-hazard industries to publish their injury and illness data annually. Both Waymo and Zoox appear because they self-classify as taxi, limousine, or transit services, the same higher-hazard category that covers a New York City livery fleet or a municipal bus depot. That classification triggered the establishment-by-establishment disclosures, including Waymo's contractor depots and Zoox's own operations, that made the worker-injury count legible at all.

Smaller test programs, startup operations, newer-city rollouts, and contractor-only depots that employ many of the same kinds of safety operators are not in the data. Absence in the OSHA record is not proof of absence in the cab.

Zoox declined to answer specific questions about the reported injuries or the continuing pattern. In a statement to TechCrunch, the company said it follows OSHA reporting requirements, takes injuries seriously, and that hard braking is sometimes unavoidable for an autonomous system.

The 2026 numbers are not yet on file. OSHA's annual establishment-specific data release, published in April 2025 for the prior calendar year, lags submissions by about a year. The 2025 baseline now in the public record is the most recent snapshot of the human operator layer of the robotaxi industry. The 2026 cycle has not closed.

Waymo and Zoox are visibly scaling operations into more cities. Each new city means more test drivers in cabs, more safety operators behind the wheel of vehicles running autonomous software that is not yet reliable enough to run without them, and more workers absorbing the mechanical surprises of a system that is still learning. The published count, by design, will not catch up to that expansion for another 12 to 18 months.

What protocols and transparency mechanisms should govern the human operator layer as that scaling accelerates? The contractor who felt the seatbelt catch her forward this summer is, in the official record, an entry in an OSHA filing at an establishment self-classified as a taxi service. The actual employer, the operating company, the city in which the test run took place, and the software that decided to brake are all in different parts of that paper trail.

The 20-plus injuries on the public record are real. The number is also a floor. The disclosure regime, the self-classification rule, the 100-employee threshold, and the annual cycle together decide which workers regulators can see. The rest of the labor that makes robotaxi scaling possible is, for now, in a part of the industry the data does not reach.