Tesla's FSD Safety Claims Don't Hold Up — And Its Own AI Trainers Know It

Tesla has repeatedly claimed its Full Self-Driving software is up to 10 times safer than a human driver — a figure cited by CFO Vaibhav Taneja, board chair Robyn Denholm, and Elon Musk himself at a shareholder meeting where Musk suggested drivers might soon be able to text while FSD handles the wheel. But a detailed examination of Tesla's statistical methodology, backed by 11 independent traffic-safety researchers, reveals the company's safety numbers are built on a series of flawed and misleading comparisons.

The most damaging error: Tesla compares its own crash rate — counting only incidents severe enough to deploy airbags — against a federal crash database that includes far less severe accidents requiring only a tow truck. That single apples-to-oranges swap inflates the safety ratio by roughly a factor of three. When University of Michigan researcher Marco Benedetti corrected for that flaw alone, the claimed 10x advantage collapsed to roughly 3x — and that's before accounting for other distortions. Tesla also benchmarks its fleet, which averages just 4.1 years old, against the entire U.S. vehicle fleet, which averages 12.8 years. Since virtually all modern vehicles now include automatic emergency braking and blind-spot monitoring that dramatically reduce crash rates, the comparison is structurally rigged. Carnegie Mellon autonomous-vehicle safety expert Phil Koopman compared it to boasting a jet is faster than a World War II bomber. Tesla further shrinks its reported crash count by only including wrecks occurring within five seconds of FSD being switched off — one-sixth the 30-second window the federal government requires automakers to use.

Behind the marketing claims, nine former data labelers and one former Tesla self-driving engineer described an FSD system that routinely struggles with basic, safety-critical tasks. Workers at a Utah labeling facility — reviewing footage from Tesla's eight-camera fleet — regularly encountered clips of FSD-equipped cars failing to yield to emergency vehicles, not stopping for school buses, speeding, missing pedestrians in crosswalks, and in at least one case nearly driving into a construction zone and striking workers. A specialized team in Palo Alto, informally called the "trauma team," was tasked specifically with reviewing near-misses involving pedestrians; engineers tightly controlled access to that footage. Seven of the nine former labelers said they personally would not trust FSD to drive them, and one said he wouldn't ride in a Tesla robotaxi "if you fucking paid me." A veteran engineer who reviewed Tesla crash data for years called the safety claims "bullshit."

The gap between public promise and internal reality is especially stark around Tesla's Austin robotaxi pilot, launched in June 2025. Former employees told reporters that in the months before the launch, Tesla workers spent enormous effort manually mapping the robotaxi zone and training the software on specific local hazards — directly contradicting Musk's repeated assertion that FSD is designed to work anywhere in real time without localized mapping, the same labor-intensive approach used by rival Waymo. Four former employees said those preparations are impossible to replicate at scale, undermining the "hyperexponential" expansion Musk has promised investors. Six months after Musk predicted FSD would cover half the U.S. population by end of 2025 and suggested texting while driving would soon be approved, neither has materialized, and Tesla's own website still warns that FSD requires active driver supervision.

The stakes are enormous. Tesla's $1.6 trillion market valuation is significantly built on the autonomous-vehicle promise. The company faces four active federal NHTSA investigations into FSD and Autopilot, including probes into vehicles running red lights, turning into oncoming traffic, and failing in low-visibility conditions — one of which involved a fatality. A Florida jury last year returned a $243 million verdict against Tesla after an Autopilot-involved crash killed a 22-year-old woman. The FTC, which NHTSA referred questions to, has brought no enforcement action. Tesla did not respond to detailed questions for the story.