Tesla AI self-driving will be >10X safer than human driving
Line chart (Q3 2018–Q1 2024) comparing miles driven per accident for Tesla with Autopilot, Tesla without Autopilot, and the U.S. average; it shows Tesla with Autopilot at 7.63 million miles per accident in Q1 2024 versus the U.S. average ~0.67 million miles, illustrating a >10× difference on a per‑mile basis.
Source: InsideEVs (Motor1 / Motorsport Network)
Research Brief
What our analysis found
Tesla CEO Elon Musk has long championed the claim that Tesla's AI-powered self-driving technology will be more than 10 times safer than human driving. Tesla's own Vehicle Safety Reports provide the data behind this assertion, with the company reporting one crash for every 7.63 million miles driven with Autopilot engaged in Q1 2024, compared to a U.S. national average of roughly one crash every 702,000 miles. The most recent Q3 2025 report showed one crash per 6.36 million miles with Autopilot active, which Tesla frames as approximately 9 times safer than the national baseline — approaching but not quite clearing the "10×" threshold in its latest quarter.
However, independent safety researchers and analysts have raised significant methodological concerns about these comparisons. Tesla's collision definition relies on airbag deployment or Delta-V thresholds under federal crash recorder standards, which excludes many lower-severity property-damage crashes captured in national statistics. Additionally, critics point out a major selection bias: Autopilot and FSD miles are disproportionately driven on highways and in favorable conditions, while the U.S. average encompasses all roads, weather conditions, older vehicles, and higher-risk urban environments — making the comparison, in the words of multiple analysts, an apples-to-oranges exercise.
Academic scrutiny has also intensified. A peer-reviewed analysis by researcher Noah Goodall, published via arXiv and Traffic Safety Research, documented revisions and reclassifications in Tesla's historical crash data that complicate independent verification. Investigative reporting in 2024 and 2025 flagged potential anomalies in how crashes are attributed when Autopilot disengages moments before impact, raising questions about whether some collisions are systematically excluded from Tesla's Autopilot tallies.
Fact Check
Evidence from both sides
Supporting Evidence
Tesla's own quarterly safety data
Tesla's Vehicle Safety Reports consistently show dramatically higher miles-per-crash figures for Autopilot-engaged driving compared to the U.S. average. In Q1 2024, the company reported one crash per 7.63 million miles with Autopilot, and in Q3 2025 reported one crash per 6.36 million miles — yielding ratios of roughly 9× to 11× safer than the national baseline of approximately 702,000 miles per crash.
Wide reproduction of Tesla's arithmetic by industry outlets
Multiple news organizations including Electrek, Teslarati, and Benzinga have independently verified Tesla's published numbers and reproduced the same ratio calculation, consistently arriving at the approximately 9–11× safety improvement that underpins the claim.
Transparent methodology disclosure
Tesla publicly documents its methodology on its Vehicle Safety Report page, including how it defines collisions under 49 C.F.R. §563.5, how it counts Autopilot-active crashes within a 5-second window before impact, and how it derives U.S. baseline figures from NHTSA and FHWA data — allowing external parties to scrutinize the approach.
Contradicting Evidence
Highway selection bias inflates the comparison
Independent safety analysts and reporters, including those at Forbes, have emphasized that Autopilot and FSD miles are overwhelmingly accumulated on highways and in favorable driving conditions, while the U.S. national average includes all road types, older vehicles, and riskier urban environments. Since highway driving already has materially lower crash rates per mile, the comparison is fundamentally skewed in Tesla's favor.
Narrow crash definition excludes many real-world collisions
Tesla counts only crashes meeting airbag deployment or Delta-V thresholds under federal crash recorder standards. This excludes a large number of property-damage-only and lower-severity collisions that are captured in NHTSA's CRSS national estimates, artificially boosting Tesla's miles-per-crash metric relative to an all-crash baseline.
Academic analysis found data revisions and classification biases
Researcher Noah Goodall's peer-reviewed analysis documented that Tesla has revised historical crash classifications and changed counting rules across reporting periods, introducing biases that make longitudinal comparisons unreliable and complicate any independent attempt to validate the 10× claim.
Disengagement-before-impact reporting gaps
Investigative analysts have flagged that when Autopilot disengages in the moments before a collision — often because the system hands control back to the driver — the resulting crash may not be counted as an Autopilot crash under Tesla's 5-second rule, potentially undercounting incidents attributable to the automated system.
Tesla itself acknowledges methodological limitations
Tesla's own methodology FAQ warns that assumptions and adjustments in comparing fleet telemetry data to federal sampling estimates can introduce biases, and that differences in highway versus non-highway mapping and reporting standards limit the validity of direct comparisons.
Report an Issue
Found something wrong with this article? Let us know and we'll look into it.