2026-06-18 — views

Physical AI Safety Benchmark — Waymo Driverless Incident Rate vs Tesla FSD Disengagement Data: What the Numbers Show

Waymo has a clean commercial driverless safety record. Tesla FSD safety data is from supervised operation only — the datasets are not directly comparable.

Overview

Safety performance is the most consequential metric in the Physical AI race. It determines regulatory approvals, public trust, insurance pricing, and ultimately commercial viability. This article benchmarks available safety disclosures from Waymo and Tesla — explaining what the data shows, how to interpret it, and what it does and does not tell investors and regulators. This is article 166 in the Physical AI Benchmark Series.

Section 1 — Waymo Safety Disclosures: What the Data Shows

Waymo operates a commercial fully driverless ride-hail fleet — no safety driver — in San Francisco, Los Angeles, and Phoenix/greater Chandler metro. Its safety record is the most closely watched in the AV industry.

Safety metric	Waymo data	Context
NHTSA SGO incident reports	Waymo files hundreds of SGO reports; the vast majority involve minor incidents (low-speed contact, property damage) rather than serious injury	SGO reports count all incidents including those where the AV was not at fault; a Waymo vehicle being rear-ended by a human driver counts as an SGO report for Waymo
California DMV collision reports	Waymo’s CA collision rate per mile driven is among the lowest disclosed by any AV operator; these reports are public and the most standardized dataset for cross-operator AV comparison	CA DMV data is the most standardized public AV safety dataset; Waymo has consistently shown low collision rates in this dataset
Serious injury or fatality incidents	No Waymo-caused fatalities in commercial driverless operations as of mid-2026 (to public knowledge); minor injuries have been reported	This is the single most important safety metric; Waymo’s commercial driverless operations have maintained a zero-fatality record commercially
Airbag deployment incidents	NHTSA SGO requires reporting of airbag deployments; Waymo’s airbag deployment rate per mile is a proxy for serious collision severity; Waymo’s rate compares favorably to human driver benchmarks at comparable miles driven	Lower airbag deployment rate equals lower rate of serious collisions
Remote assistance requests (vs disengagements)	Waymo’s commercial fleet operates fully driverlessly; “disengagements” do not apply; instead, Waymo tracks remote assistance requests where a remote operator helps navigate an unusual scenario	Remote assistance request rate is Waymo’s equivalent of a disengagement rate; specific rates are not publicly disclosed
Comparison to human driver baseline	NHTSA data: US human drivers have approximately 1.35 fatalities per 100M miles driven and approximately 5–6 injury crashes per million miles; Waymo’s disclosed incident rates compare favorably to these baselines	This comparison is imperfect: Waymo operates in geofenced urban areas, avoids highway miles, and mostly operates in good weather — a direct per-mile comparison has meaningful ODD caveats
Safety culture indicators	Waymo publishes an annual Safety Report with detailed methodology, incident categorization, and simulation test results	Published safety methodology signals regulatory maturity; Waymo’s Safety Report is considered a benchmark for AV safety disclosure transparency

Section 2 — Tesla FSD Safety Disclosures: What the Data Shows

Tesla’s fleet scale is unmatched — approximately 6 million vehicles with Autopilot or FSD engaged across hundreds of millions of trips. The safety data this generates is extensive but comes from an entirely different operational regime than Waymo’s.

Safety metric	Tesla data	Context
NHTSA SGO incident reports (Tesla Autopilot/FSD)	Tesla has the highest raw number of SGO reports among AV operators because it has the largest fleet engaging ADAS/AV features (~6M vehicles)	Raw SGO count is misleading without per-mile normalization; Tesla’s high count reflects fleet scale, not necessarily higher per-mile risk
NHTSA investigations	NHTSA has opened multiple investigations into Tesla Autopilot involving emergency vehicle collisions, crashes in construction zones, and phantom braking incidents; some were closed without regulatory action, others led to software updates	Investigations do not equal fault findings; Tesla’s software update response to NHTSA findings has generally been accepted as adequate
Tesla quarterly safety report	Tesla publishes quarterly Autopilot/FSD safety data: miles per airbag deployment and miles per crash; Q4 2025 (est.): Tesla reported one crash per 7.14M miles with FSD engaged vs. US average of one crash per ~702K miles	The 10x+ safety improvement claim is compelling but contested: Tesla’s fleet drives many highway miles, which are inherently safer than urban intersections regardless of driver
Critical disengagement rate	FSD is supervised — humans disengage when uncomfortable; Tesla tracks critical disengagement rate (driver intervention that averted a likely collision) per 1,000 miles; Q4 2025 (est.) rate declined significantly YoY under FSD v13	Critical disengagement rate is a proxy for how often FSD makes an error serious enough to require human correction; declining rate signals improving model quality
Austin Robotaxi safety	Tesla’s Austin Robotaxi launched in 2026 with a safety driver present initially; transition to fully driverless requires NHTSA FMVSS exemption for Cybercab (no pedals/steering wheel)	Tesla’s driverless safety record is not yet established because true driverless commercial operation has not yet begun at scale
Serious injury or fatality incidents	Tesla Autopilot/FSD has been involved in fatal accidents; investigations have produced mixed findings on whether Autopilot was a contributing factor vs. driver negligence or misuse	The presence of fatalities in Autopilot-engaged operation does not automatically mean Autopilot was at fault; NHTSA investigations determine causal responsibility; this remains a contested area
Key distinction: supervised vs. driverless	All Tesla FSD safety data is from supervised operation (human driver present and responsible); Waymo’s commercial data is from fully driverless operation (no human driver)	The supervised/driverless distinction is the most important context for comparing Tesla and Waymo safety data; they are not directly comparable

Section 3 — How to Interpret Safety Data: The Comparison Problem

A naive per-mile comparison between Waymo and Tesla safety data is misleading. Five structural differences make direct comparison problematic.

Comparison challenge	Detail	What it means for investors
Operational design domain (ODD) mismatch	Waymo operates in geofenced urban areas at low speeds (max ~65 mph); Tesla FSD operates on highways, urban streets, and rural roads across diverse weather conditions	A Waymo mile in downtown San Francisco is categorically more complex than a Tesla FSD mile on Interstate 5 in clear weather; per-mile comparisons without ODD normalization are misleading
Supervised vs. driverless operation	Tesla FSD has a human driver available to intervene at any moment; Waymo commercial driverless has no such fallback; a Waymo error is always an AV error; a Tesla FSD near-miss may be a human save	Waymo’s safety record is measured under higher-consequence conditions than Tesla FSD; the fallback asymmetry matters for both safety interpretation and regulatory treatment
Reporting standards vary	NHTSA SGO counts incidents differently for AV operators vs. ADAS; CA DMV reporting captures CA operations only; Tesla’s quarterly reports use Tesla’s own methodology; no unified standard exists	Without a unified safety reporting standard, all cross-company comparisons are approximate; investors should be skeptical of “X is safer than Y” claims that do not specify the methodology
Scale affects statistics	Waymo has approximately 30M driverless commercial miles (est.); Tesla has approximately 6B FSD-supervised miles (est.); at 30M miles, a single serious incident produces a very different per-mile rate than at 6B miles	Waymo’s safety record looks excellent partly because the statistical sample is small; as Waymo accumulates more miles, the probability of at least one serious incident increases — this is statistical normalization, not deteriorating safety
What independent analysts track	RAND Corporation has estimated that AV companies need billions of miles (not millions) to statistically prove superiority to human drivers at high confidence; at current scale, neither Waymo nor Tesla has enough driverless miles for definitive statistical comparison	”No incidents so far” at 30M driverless miles is promising but not statistically conclusive proof that AVs are safer than humans at the societal scale needed for broad deployment

Section 4 — Regulatory Safety Framework: What Determines Commercial Approvals

Regulatory approval is the gatekeeping function in the Physical AI race. The key approvals and their status differ materially between the two companies.

Regulatory dimension	Waymo status	Tesla status	What it means
California CPUC driverless permit	Waymo holds CPUC permits for commercial driverless ride-hail in San Francisco and Los Angeles; permits require ongoing safety reporting and can be suspended for cause (as happened with Cruise in 2023)	Not applicable; Tesla has not applied for a CPUC driverless permit (no driverless vehicle currently in CA commercial service)	Waymo’s CPUC permit is the gold standard of US AV regulatory approval; Cruise’s 2023 suspension shows permits can be revoked if safety events warrant
NHTSA FMVSS exemption	Not required; Waymo’s vehicles have steering wheels and standard safety equipment meeting FMVSS	Required for Cybercab (no steering wheel, no pedals); Tesla has applied for FMVSS exemption; outcome pending as of mid-2026	FMVSS exemption is the key regulatory bottleneck for Tesla’s Cybercab commercial driverless deployment; denial would require redesigning the vehicle
Arizona DOT permits	Waymo operates in Phoenix/Chandler/Tempe/Scottsdale under Arizona DOT framework; AZ has the most permissive AV regulations in the US	Not applicable for driverless; Tesla operates supervised FSD in AZ	Arizona’s permissive regulatory environment is a structural advantage for Waymo’s most mature market
Texas (Austin) regulatory environment	Waymo operates in Austin under Texas AV framework (permissive)	Tesla launched Austin Robotaxi under Texas framework; driverless permit for Cybercab requires FMVSS exemption path	Both companies benefit from Texas’s permissive AV regulations; Texas does not require specific state AV permits beyond FMVSS compliance
National Highway Safety standards	Waymo complies with all applicable FMVSS; safety reports filed regularly	Tesla Autopilot/FSD subject to ongoing NHTSA oversight; multiple investigations; Cybercab FMVSS exemption pending	NHTSA’s treatment of the Cybercab FMVSS exemption is the single most consequential regulatory decision for Tesla’s Physical AI timeline in H2 2026

Section 5 — Safety Benchmark Scorecard

Safety dimension	Waymo	Tesla FSD	Edge	2028 outlook
Driverless commercial safety record	Clean (no commercial driverless fatalities, est.); extensive driverless miles under CPUC oversight	No commercial driverless operation yet (Austin is supervised)	Waymo (established driverless record)	Tesla begins building a driverless record as Austin transitions to Cybercab
Per-mile incident rate (supervised context)	Low CA DMV collision rate; favorable NHTSA SGO per-mile rates	Favorable quarterly reports (est. 10x+ vs. US average); methodology contested	Both favorable vs. human baseline; not directly comparable	Both improve as AI models improve
Regulatory approval status	CPUC driverless permits (CA); AZ and TX frameworks; most comprehensive approval set	FMVSS exemption pending for Cybercab; no CPUC driverless permit; supervised operation approved broadly	Waymo decisive (current regulatory approvals)	Tesla FMVSS exemption outcome is the key 2026 catalyst
Safety transparency	Annual Safety Report; detailed methodology; CA DMV public data	Quarterly safety reports; NHTSA SGO filings; less granular than Waymo disclosure	Waymo (more transparent)	Industry pressure for standardized disclosure increasing
Incident severity history	Minor injuries; no commercial driverless fatalities (est.)	Autopilot/FSD-involved fatalities in supervised operation (causation contested)	Waymo (no driverless fatalities)	Tesla driverless record starts fresh when Cybercab launches
Overall safety verdict	Waymo holds the stronger safety record on the dimensions that matter most for regulatory approval: driverless operation (no human fallback), CPUC permit maintenance, and transparent reporting. Tesla’s safety data is extensive but comes from supervised operation where human fallback is always available. The fairest comparison is “Waymo driverless safety” vs. “Tesla will establish driverless safety when Cybercab launches” — and that comparison does not yet have enough Tesla driverless data to score. For investors, Waymo’s safety track record is a current moat; Tesla’s safety track record as a driverless operator is a future unknown.

All figures labeled (est.) are derived from public company disclosures, analyst estimates, and industry benchmarks. This article is part of the Physical AI Benchmark Series — article 166.