A safety report starts with definitions, not the headline number
Uber’s U.S. safety reports and Lyft’s transparency reports cover serious incidents connected to their platforms, including sexual assault, fatal physical assaults and fatal motor-vehicle crashes. That sounds comparable at first glance. The problem is that the details matter: which years are included, which categories are counted, how incidents are classified and what denominator is used.
Lyft’s July 2024 report, for example, covered 2020 through 2022 and focused on defined categories of serious safety incidents. Lyft said the incidents included in that report occurred in a very small fraction of rides. Uber’s reports also use defined categories and explain their methodology. A percentage drawn from one report should not be compared mechanically with a percentage from another unless the underlying definitions match.
Sexual-assault categories are especially easy to misread
Safety reports do not simply contain one universally defined number called “sexual assaults.” Companies may group conduct using a taxonomy that distinguishes categories by the type of contact or conduct reported. A report can therefore appear to show a dramatic difference depending on whether a writer uses only the most serious categories or includes a broader range of sexual misconduct.
That is one reason the Government Accountability Office did not treat the public company reports as a complete national assault database. A company can accurately report what falls within its own methodology while the public still lacks a uniform cross-company dataset.
The year ranges do not always move together
Another problem is timing. GAO reported in 2026 that comparable public rideshare assault counts for 2023 through 2025 were not yet available from the selected companies as of June 2026. Lyft has separately said its next safety report will cover 2023 through 2025 and that it plans broader disclosure across sexual-assault and misconduct categories.
Until reporting cycles line up, a current-looking webpage can contain data ending several years earlier. That does not make the report useless. It does mean a reader should check the covered period before treating the numbers as a description of present-day risk.
The denominator can change the story too
Ride counts are often used to show how rare a category of incident is. That is legitimate context, but denominators deserve the same scrutiny as incident counts. A rate “per trip” answers a different question from a rate per passenger-mile, per active user or per hour of exposure.
Large platforms also complete enormous numbers of trips. A low per-trip rate can coexist with a substantial absolute number of reported incidents. Both facts can be true at the same time.
Company reports are useful — just not independent audits of the entire industry
Uber and Lyft safety reports provide information that would otherwise be difficult to obtain. They describe incidents reported to the companies, explain safety programs and create a public record that can be compared over time within the limits of each methodology.
But they remain company-produced reports. GAO’s 2026 review is valuable precisely because it places those disclosures beside police data, federal information and other sources and then identifies what is still missing.
The responsible comparison is narrower than “Which app is safer?”
A defensible Uber-versus-Lyft safety comparison should start by matching the same years and the same incident definitions. It should state whether the number comes from company reports, police data or another source. And where the reporting does not line up, it should say so instead of manufacturing a winner.
That answer may be less satisfying than a simple ranking. It is also much closer to what the public evidence can actually support.
Sources
