FSD Safety Data Validity
FSD Safety Data Validity
One-line summary: Tesla's public FSD safety metrics have multiple methodological issues that, per independent analysis, make them unfit to support the company's "safer than humans" claims.
The insight
The question "how safe is tesla-fsd really?" can't currently be answered from public data. Tesla publishes a safety report, but independent researchers — most visibly Phil Koopman — argue the methodology is biased enough that no meaningful practical-safety conclusion can be drawn from it.
Evidence
From Koopman's analysis of Tesla's public FSD safety report 2026-04-20-autoresearch-tesla-fsd:
- Flawed baseline. Tesla compares new FSD-equipped vehicles against 12+ year-old cars lacking modern safety features — not against modern ADAS-equipped peers.
- Reporting window mismatch. Tesla uses a 5-second crash-attribution window; nhtsa's standard is 30 seconds. Crashes where FSD contributed but handed off >5s before impact are excluded.
- Under-reporting of severe crashes. Tesla has "systemic under-reporting bias for the most severe crashes" because telematics can be disabled by heavy damage, including fatalities.
- Selective crash categorization. Excludes crashes without airbag deployment in the FSD numerator while the baseline includes such incidents.
- Highway-definition ambiguity. Tesla's undefined "highway" classification could mask up to ~5.7× difference against actual NHTSA fatality data.
Undisclosed data. Tesla does not publish disengagement data, crash-by-severity, miles by road type, or the denominator it uses for its crash-rate calculation 2026-04-20-autoresearch-tesla-fsd.
Community-tracker trend. Third-party FSD community-tracker "city miles to critical disengagement" fell from ~4,109 (v14.1 peak, October 2025) to ~809 after v14.2 — a sharp regression if the metric is reliable 2026-04-20-autoresearch-tesla-fsd.
Koopman's conclusion: "The threats to validity are so pervasive… we can conclude nothing useful about the practical safety of FSD from this report." IIHS classifies FSD as a convenience feature, not a safety improvement 2026-04-20-autoresearch-tesla-fsd.
Design implications
Treat all Tesla-reported FSD safety numbers as lower-confidence than comparable waymo reporting (which is per-trip and regulator-visible). For safety claims, cite independent analyses rather than Tesla's report directly.
Contradictions / tensions
- Musk publicly claims FSD is "10× safer than humans." Koopman and sellside analysts (citing community-tracker data) both conclude the opposite is at least plausible; at minimum, the claim is unsupported 2026-04-20-autoresearch-tesla-fsd.
Open questions
- Will Tesla publish denominator and crash-by-severity data under regulatory pressure?
- Does nhtsa EA26002 force methodology disclosure?