Inside the Reuters Investigation That Dismantled Tesla’s ‘10X Safer’ FSD Claim

Tesla told regulators that vehicles using FSD travel more than seven times farther between crashes than vehicles driven by humans, pegging the system as up to 10 times safer than the average human driver . One presentation slide went further, claiming that FSD could have “saved 32,000 lives” and prevented 1.9 million injuries in the U.S. alone . That figure was derived by assuming every single vehicle on American roads—from motorcycles to freight trucks—was replaced by an FSD-enabled Tesla, an assumption independent researchers called “absurd” .

The Flawed Comparison at the Core of the Data

The methodological errors identified by Reuters are both fundamental and deeply distorting .

The wrong baseline. Tesla did not compare its FSD-driven miles to other vehicles of a similar age and technological capability. Instead, it benchmarked its modern, crash-avoidance-equipped fleet against the National Highway Traffic Safety Administration’s (NHTSA) average for the entire U.S. fleet, which includes vehicles over a decade old with far fewer safety features. This stacked the deck in Tesla’s favor before the analysis even began .

Cherry-picked incident severity. The most consequential error was in how crashes were counted. On its own side, Tesla counted only the most severe subset of crashes—those where an airbag deployed. For the human-driver baseline, however, it used a much broader federal data category that includes all crashes serious enough to require a tow truck. A tow-truck crash is a far less severe threshold; it often includes low-speed parking lot collisions and minor fender-benders where no airbag ever deploys .

The U.S. federal database already breaks out airbag-deployment crashes as a separate, directly comparable metric. Tesla chose not to use it. Seven of nine independent researchers who reviewed the figures told Reuters this single decision inflated Tesla’s claimed safety advantage by roughly a factor of three . In other words, an apples-to-apples airbag-to-airbag comparison would have slashed Tesla’s headline “10X safer” claim to something significantly lower .

Self-published and unverified. All of the statistics were self-reported by Tesla. Reuters confirmed the data had never been independently audited or verified by an outside body, meaning regulators were effectively asked to take the company’s marketing claims at face value .

The People Who Train the AI Don’t Trust It

The Reuters investigation didn’t stop at the numbers. In the May 28 report, journalists spoke with more than nine former Tesla data labelers—the workers who spend hours training the FSD AI to recognize objects and scenarios on the road—as well as a former self-driving engineer .

These employees expressed deep personal skepticism about the technology. Seven of nine former data labelers told Reuters they would not trust FSD to drive them . They cited regular failures around emergency vehicles, school buses, motorcyclists, off-ramps, and construction zones, and described spending extensive hours training the system under specific, ideal conditions to boost its performance in demos . Their lack of confidence in the technology they helped build undercut the polished safety statistics the company was presenting to European officials.

Regulators and Researchers Respond

The reaction from the independent research community was nearly unanimous. Reuters interviewed 11 traffic-safety experts to evaluate Tesla’s methodology. Ten of them concluded that what Tesla had presented was not a valid safety analysis—it was a marketing document dressed up in the language of science .

The experts told Reuters the data could not be used to credibly support FSD’s safety claims, and several characterized the approach as deliberately misleading . The Swedish Transport Agency, one of the bodies that received the presentation, emphasized that regulators must look beyond surface-level numbers when evaluating safety claims .

As of the June 2026 reports, European regulators had not issued a definitive public ruling rejecting Tesla's data. However, the company had already secured limited approval for FSD in parts of Europe, and the Reuters investigation raised pointed questions about whether that approval was obtained on the back of inflated statistics .

The investigation makes clear that the most widely publicized safety claim in the autonomous driving industry was supported not by rigorous, independently-verified science, but by a comparison methodology that multiple experts said was engineered to produce a favorable headline.