Product1 distinct publisher3 min readPublished
A TechCrunch review of OSHA filings counts more than two dozen injuries to Waymo and Zoox test drivers across 2024 and 2025. The figures are public at all only because both companies file as higher-hazard taxi operators.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
The wrist is the detail worth sitting with. A Waymo test driver reported that the car suddenly backed up while parking itself and whipped the steering wheel, catching his left hand and bending it until he heard a crack; he was away from work for more than two months [16]. The same event lands in two places at once, as a note about low-speed reverse behaviour in a validation log and as an injured hand in a federal filing. OSHA data and worker interviews also record sprains, strains and whiplash, with some drivers sidelined for weeks or months [4].
The counts narrow where the problem sits. Transdev logged five Waymo test-driver injuries in 2024 across San Francisco, Los Angeles and Phoenix, one from hard braking and four from erratic vehicle behaviour [14], then 11 in the same three markets in 2025, with hard braking the cause of nearly all of them [15]. Five plus 11 is the 16 that Transdev reported in total [2], so the public record for Waymo covers three cities and no others [1]. Year over year that is a 2.2x rise [2]. Waymo's commercial fleet grew from more than 1,500 vehicles in May 2025 to more than 3,500 as of TechCrunch's August 2026 report [18], roughly 2.3x over a partly overlapping window [3]. Waymo does not break out how many vehicles it runs for testing [19], so nobody outside the company can convert any of this into a rate.
Zoox told TechCrunch that hard-braking incidents are sometimes unavoidable, and that the injuries it reported last year represent a small percentage of the millions of miles its test fleet travelled [11]. Miles are the denominator for a vehicle. A test driver occupies one seat for a shift and absorbs whatever that seat does. The Phoenix driver who spent 157 days away from work after the car made an exaggerated braking event with no obstruction in front of it [17] lost about five months [4], and dividing that by fleet mileage does not make it shorter for him.
The visibility is an artifact of paperwork. Publication reaches only establishments with 100 or more employees in these industries, so drivers hurt in newer cities where operations have not scaled would not appear [6]. Other autonomous vehicle companies are absent from the dataset altogether, though TechCrunch notes they may be exempt [7]. The 2026 figures are not due until the following year under OSHA's Injury Tracking Application, and TechCrunch could not learn whether Waymo or Transdev logged anything new this year [8]. Zoox declined to answer specific questions, saying it follows the necessary reporting requirements and takes injuries seriously [10]. Waymo also declined, pointing to tens of millions of miles driven with human operators as part of its validation process [12], and Transdev, which employs and manages those drivers, declined to comment [13].
If you run humans inside the loop of any system, the useful instrument is a reconciliation rather than a dashboard. Keep two columns for every hard brake or sudden manoeuvre above a threshold you pick: what the stack logged, and what the person in the seat said afterwards. Column one with nothing in column two is ordinary. Column two with nothing in column one is a hole in your telemetry, and it is the expensive kind. Then report the result per seat-hour, because that is the unit the injury actually happens in. The tradeoff is real: reconciling will make your event counts look worse than any per-mile figure, and collecting it costs shift time from people you are already asking to sit through phantom stops. The alternative is that your first dependable signal about worker harm is a filing your staffing vendor makes a year later, which is more or less where this story came from.
Ranked by verification strength, evidence, and original report placement.
According to a TechCrunch review of data submitted to OSHA, test drivers for Waymo and Zoox sustained more than two dozen injuries in 2024 and 2025 from hard braking or other sudden movements made by the autonomous vehicles.
Transdev, which employs and manages Waymo's test drivers, reported 16 injuries tied to Waymo depot locations in three cities that were caused by the autonomous software braking hard or swerving suddenly.
Zoox reported as many as eight worker injuries linked to hard braking.
Test drivers have suffered sprains, strains, whiplash and more, according to OSHA data and interviews with former and current workers, and some have been sidelined for weeks or months.
The Zoox and Waymo injuries appear in the dataset because the companies classify themselves as taxi, limo or transit services, which OSHA subjects to greater disclosure as "higher-hazard" industries.
OSHA requires only establishments with 100 or more employees in these industries to make this information public, meaning test drivers in newer cities where operations have not scaled may have been injured on the job without appearing in the dataset.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 27, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
DC wants 15 cents a mile from robotaxis. That is a price, not a permit.1 distinct publisher
product
Waymo's 5nm ingest chip moves autonomy economics into the silicon design budget1 distinct publisher
product
Waymo puts its 200 million driverless miles up against the Cybercab's cameras1 distinct publisher
product
OpenAI's plan to hand everyone a coding agent leaves the hard part to the model1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Primary filings, single publisher
The core counts rest on regulatory filings submitted by Transdev and Zoox to OSHA, quoted with dates, cities, causes and lost-workday figures, and are reinforced by interviews with two current and three former Zoox contractors. That is unusually hard sourcing. It is discounted because the cluster contains exactly one publisher with no independent corroboration, the worker accounts are anonymous, all three companies declined to address specifics, and the article itself flags that the dataset is partial (higher-hazard classification only, 100-employee threshold, one-year reporting lag, other AV firms absent).
Scaling fleets, undisclosed test denominators
Real-world deployment behind the story is substantial and dated: Waymo's commercial fleet went from more than 1,500 vehicles in May 2025 to more than 3,500 by August 2026, with testing in three disclosed injury-reporting markets plus new cities, and Zoox running a modified Toyota Highlander test fleet reported active as recently as July. Adoption is not scored higher because neither company breaks out testing fleet size, test-driver headcount or test mileage, so the exposure base cannot be sized from this cluster.
Likely understated by disclosure limits
The reporting's own framing is more conservative than its evidence supports. 'More than two dozen' is a floor, not a total: only self-classified higher-hazard taxi/limo/transit establishments file publicly, only sites with 100 or more employees must publish, newer cities are therefore invisible, 2026 data is not yet due, and no other AV company appears at all. Company framing pushes the other way -- Zoox calls the injuries a small percentage of millions of test miles without publishing the denominator -- but the structural undercount dominates, so the story reads understated rather than overstated.
Strong disclosure asymmetry
Every party with primary knowledge has an incentive to say less: Waymo and Zoox both declined specifics and redirected to safety-mission and aggregate-mileage framing, and Transdev -- the employer that actually filed the 16 reports -- declined to comment entirely. Workers spoke only under anonymity to discuss private operations, indicating perceived retaliation or contractual risk. The publisher has a competitive incentive toward an exclusive investigative finding, which is visible in the framing but constrained by the use of quoted regulatory filings.
Solid on filings, thin on totals
High confidence in the specific documented incidents, counts, markets and lost-workday figures, because they come from regulatory filings quoted directly. Moderate overall because the cluster has a single publisher, the totals are structurally incomplete, continuation of harm rests on five anonymous accounts, no regulator or independent expert is quoted, and no normalised injury rate is obtainable without undisclosed test-fleet denominators.