Autonomous vehicles are simultaneously the best-documented and most contested safety story in physical AI — Waymo has logged over 220 million fully driverless miles with a peer-reviewed safety record that beats human drivers by a wide margin, while Tesla's much smaller robotaxi fleet is posting crash rates that multiple independent analyses put well above the human baseline. Both of those things are true at the same time, and the gap between them says more about deployment discipline and transparency than about whether autonomous driving works in principle.
This is the latest installment in our ongoing series looking past the headlines at what the actual data shows — following earlier pieces on AI's social impact, AI in healthcare, and physical AI in manufacturing and care settings. As always, treat the figures below as a mid-2026 snapshot in a field that is changing month to month.
Quick Comparison: Where the Evidence Actually Stands
| Domain | What the Data Shows | Where the Risk Concentrates | The Honest Takeaway |
|---|---|---|---|
| Safety Record | Waymo: 13x fewer serious-injury crashes than human drivers over 220M+ driverless miles | Tesla's robotaxi fleet reportedly crashes at a rate several times higher than the human baseline, even with a safety monitor onboard | Autonomy itself isn't the variable — fleet design, scale, and caution clearly are |
| Accountability & Transparency | Waymo publishes detailed, narrative crash reports to NHTSA | Tesla redacts nearly every crash narrative and has been investigated for delayed crash reporting | The same regulatory framework produces very different transparency depending on the company |
| Labor & Regulation | Driverless trucking logged 250,000+ miles with zero attributed collisions on mapped highway corridors | Federal liability and labor law still largely assumes a human is behind the wheel | The technology is outrunning a regulatory and legal framework built for a different era |
Safety Record: Real Gains, Real Gaps, and a Wide Spread Between Operators
Start with the strongest evidence in the entire physical AI space: Waymo's own published safety data, which has held up under independent and peer-reviewed scrutiny. Through over 220 million rider-only miles, Waymo's research concluded its vehicles are 13 times less likely to be involved in a serious injury crash than human drivers, and the company's peer-reviewed analysis estimated a typical human driver would have caused more than 500 additional injury crashes over the same distance. That's not a marketing claim sitting alone — it's been compared against NHTSA crash data and benchmarked against human driving rates across the specific cities where Waymo actually operates.
Tesla's robotaxi program tells a much rougher story so far, and the gap is stark enough that multiple independent outlets reached the same conclusion using Tesla's own disclosed mileage. One analysis calculated Tesla's robotaxi fleet was crashing roughly once every 55,000 miles with a human safety monitor present in every vehicle, compared to the roughly 229,000 to 500,000 miles between police-reported crashes for an average human driver — putting Tesla's rate at three to four times worse than human drivers, even with a person in the car who could intervene. Waymo's fully driverless fleet, by contrast, has logged over 127 million miles at an average of one crash per roughly 98,000 miles, all without any human monitor at all.
It's worth being precise about what these numbers do and don't prove. Neither company's robotaxi fleet is remotely comparable in scale — Waymo operates thousands of vehicles across multiple major cities, while Tesla's robotaxi fleet has numbered in the dozens, concentrated in a small part of one city. CNN's independent investigation also found a more nuanced safety problem hiding inside Waymo's strong aggregate numbers: hundreds of documented incidents where robotaxis ran red lights, entered active emergency zones, or made dangerous maneuvers that a human driver would have instinctively avoided — the flip side of removing human error is introducing entirely new categories of machine error that didn't exist before.
Accountability & Transparency: The Same Rules, Very Different Behavior
Both companies report to the same federal crash database, and the contrast in how they use it is itself a data point. A typical Waymo crash report reads like an actual incident description — vehicle, direction of travel, specific maneuver, what the other party did, and the outcome. Tesla's crash narratives in the same NHTSA database are consistently redacted with the same boilerplate phrase, leaving outside analysts unable to determine basic facts like who was at fault or what specifically happened in a given collision.
This isn't a one-time lapse, either. Tesla revised a July 2025 crash report nearly six months after the fact, upgrading its severity from "property damage only" to "minor injury with hospitalization" — and NHTSA has separately opened an investigation into Tesla for a pattern of late crash reporting. Whatever one concludes about the underlying safety of either company's technology, the transparency gap between them is measurable and well-documented, and it directly shapes how much outside verification of safety claims is actually possible.
Labor & Regulation: The Trucking Story Is More Contained Than the Headlines Suggest
On the freight side, where the labor stakes are arguably highest, the actual deployment numbers are far smaller than "driverless trucks are taking over the highways" coverage implies. Aurora Innovation has logged over 250,000 driverless miles with zero Aurora-attributed collisions across mapped Sun Belt corridors and is targeting roughly 200 fully driverless trucks by the end of 2026. Kodiak AI reported more than 23,500 cumulative hours of paid driverless operation as of its first-quarter 2026 results, a 120% increase over the prior quarter. Those are genuinely significant technical milestones.
But scale them against the actual size of the industry, and the picture changes. Industry analysts estimate fewer than 50 fully driverless trucks are operating commercially across all companies combined, against roughly 4 million Class 8 trucks nationally — and even optimistic forecasts put commercial Level 4 autonomy at just 5% to 10% of long-haul lanes by 2032. That's nowhere near large enough to address a driver shortage industry analysts already estimate at 60,000 drivers in 2026, projected to grow to 160,000 by 2028. For now, autonomous trucking looks far more like a narrow, highway-corridor solution than an imminent replacement for the broader trucking workforce.
The regulatory and legal framework underneath all of this is, by most informed accounts, still playing catch-up. Legal analysts note that hours-of-service rules, driver qualification standards, and most of the Federal Motor Carrier Safety Regulations were written assuming a human is physically in the vehicle — a structure that doesn't cleanly apply once that assumption is no longer true. Liability law is shifting in response: several legal analyses point to a move toward product-liability-style frameworks, where a plaintiff needs to show the vehicle or its software was defective rather than that a human driver was negligent, shifting risk toward manufacturers and software developers rather than individual operators. At the federal level, the U.S. still lacks a comprehensive AV framework, leaving a patchwork of state rules — Texas now requires AV operators to hold a state permit as of May 2026, while California's updated 2026 regulations newly empower local emergency officials to geofence robotaxis out of active emergency zones and require manufacturer responses to first responders within 30 seconds.
So What Does This Mean for Where Autonomous Vehicles Actually Go Next?
Pulling the three threads together, the clearest finding in this entire space is that "autonomous vehicles" isn't one technology with one safety profile — it's a category containing operators with dramatically different safety records, transparency practices, and deployment discipline.
- On safety, the evidence supports cautious optimism about the technology in principle, paired with real skepticism about any specific company's claims until independently verified — Waymo and Tesla are not, by the numbers, in the same category right now.
- On accountability, transparency itself should be treated as a safety signal — a company willing to publish detailed, unredacted crash narratives is giving outside observers something to actually verify; one that redacts everything is asking for trust without evidence.
- On jobs and regulation, the displacement timeline for trucking specifically is more gradual than the "robots are taking truckers' jobs" framing suggests, but the legal and regulatory structure genuinely hasn't caught up to vehicles that no longer assume a human is behind the wheel, and that gap is where the real near-term risk and confusion sits.
None of this argues that autonomous vehicles are either a solved safety triumph or a reckless gamble — the honest read is that the technology's ceiling looks genuinely high, based on the best operator's data, while the floor depends entirely on which company, which fleet, and which regulatory jurisdiction you're actually talking about.
Frequently Asked Questions
Are robotaxis actually safer than human drivers?
It depends heavily on the operator — Waymo's peer-reviewed data shows 13 times fewer serious-injury crashes than human drivers across over 220 million driverless miles, while independent analyses of Tesla's smaller robotaxi fleet found crash rates several times higher than the human baseline.
Why does Tesla redact its robotaxi crash reports?
Tesla has not publicly explained its redaction practice, but it stands in contrast to competitors like Waymo, which publish detailed narrative crash descriptions, and Tesla has separately faced an NHTSA investigation over delayed crash reporting.
Will self-driving trucks replace human truck drivers soon?
Not at meaningful scale yet — fewer than 50 fully driverless trucks operate commercially across all companies combined, against roughly 4 million Class 8 trucks nationally, and most forecasts put Level 4 autonomy at only 5% to 10% of long-haul lanes by 2032.
Who is legally liable if a self-driving car causes a crash?
Liability frameworks are actively shifting toward product-liability standards, where the manufacturer or software developer can be held responsible if the vehicle or its software is shown to be defective, rather than placing fault on a human driver as in traditional auto accident law.
Is there a federal law regulating self-driving cars in the US?
Not yet a comprehensive one — the U.S. currently relies on a patchwork of state-level regulations, though proposals like the SELF DRIVE Act aim to establish unified federal safety standards, and individual states such as Texas and California have already passed their own updated AV rules in 2026.