How Safe Is Safe Enough?

Here’s a number that should stop you mid-scroll. Independent researchers at the Insurance Institute for Highway Safety (IIHS) spent months digging through federal crash data on Waymo’s robotaxis, cleaning up the mess of duplicate reports and unreportable fender benders, and came out the other side with this: Waymo’s driverless vehicles got into 68% fewer crashes than human drivers covering the same roads.

That’s not a company bragging about itself. That’s an independent safety research group checking the homework.

Keep that number in your back pocket, because it’s not about trucks. Not yet. We’ll get to what that means for freight later.

Break it down further and it gets more interesting. Waymo vehicles rear-ended other cars 91% less often than humans do. T-bone and angle crashes — the kind that happen at intersections and kill people — dropped 61%. They ran off the road into trees zero times, something human drivers manage to do with depressing regularity. See Figure 1 for a comparison of Waymo crash rates to that of human drivers. Waymo’s own larger dataset, now well past 200 million driverless miles, backs it up. It has 94% fewer crashes involving serious injury or worse compared to humans driving the same streets.

Figure 1. Waymo crash rate by type vs. human drivers

Sit with that for a second. Two out of three crashes, gone. Nine out of 10 rear-end crashes, gone. That’s the kind of population-level shift that gets a new vaccine fast-tracked or a new intersection design copied in every city in the country. It doesn’t mean nobody gets hurt. It means fewer people do.

So why does it feel like nobody trusts this thing?

The Perfection Trap

Here’s the part that should bother you if you’re in this industry. Most people don’t hear “68% safer” and think progress. They hear it and think “so it still crashes.” That reaction isn’t irrational, and it isn’t the public being difficult for the sake of being difficult.

There’s real research behind it. A 2019 study, published in the Journal of Safety Research, found that most drivers rate their own driving as above average, then hold self-driving cars to that inflated personal standard instead of the real population average.

Researcher Michael Nees found people wouldn’t feel safe riding in, buying, or even allowing a self-driving car on public roads until it beats their own perceived skill, not just the skill of the average driver everyone already assumes is worse than they are. Since almost nobody thinks they’re an average driver, safer than average was never going to be the standard the public actually applied.

A larger 2021 study, published in Transportation Research Part C, found the same bias holds up at scale. Across more than 4,500 people, it showed the public stays resistant to self-driving cars even when the cars are demonstrably safer than the average human, unless the safety margin gets extreme.

It’s a strange bar to set, and it gets stranger the more you think about who’s setting it: us. Human drivers are, on average, not particularly safe. We speed. We text. We drive tired, drive drunk, drive distracted, and occasionally drive straight into a tree for no good reason at all. And yet we hold the technology built to fix that record to a standard that no human driver on the road could ever clear themselves.

Call it the perfection trap. We want zero errors from the machine while accepting a whole lot less than zero from ourselves.

There’s a harder version of the perfection trap too, and it’s not just psychological. The Ford Pinto passed federal standards, and the 737 MAX flew safely on the vast majority of its flights, yet both still ended in massive lawsuits and redesigns, because statistical safety and legal safety are not the same test. That gap, and what it means for freight, deserves its own piece rather than a footnote here.

Riding Changes the Math

There’s a cleaner way to see this trap in action than any lab study. Look at what happens to people once they actually get in the car.

A national AAA survey from February 2025 found that just 13% of U.S. drivers said they’d trust riding in a self-driving vehicle, barely up from 9% the year before. Six in 10 said they’re still afraid to get in one.

Now compare that to a UC Berkeley SafeTREC survey of San Francisco residents who’d actually interacted with an AV. Among people who’d ridden as a passenger, 72.5% said they trust AVs. See Figure 2. Even the human drivers who just share the road with them, never having ridden in one themselves, trusted the technology more than the national average, at 47.1%.

Figure 2. Trust in AVs: national average vs. firsthand experience

That’s not a subtle gap. Thirteen percent trust nationally. Seventy-two percent among people who’ve actually been a passenger. That’s not a shift, it’s practically a different technology in the public’s mind. What changed wasn’t a new statistic. It was the ride itself.

That’s a hard lesson for anyone trying to build public trust with a press release. You can publish every peer-reviewed study in the world and move the needle less than a single ride does.

Freight won’t get that shortcut. There’s no passenger seat for a truck to win someone over. Whatever trust freight autonomy earns, it must earn from highway drivers sharing the lane, fleet managers signing the contracts, and lawmakers writing the rules, none of whom get to feel what it’s like riding inside the vehicle. That makes the data argument carry more weight for trucks than it ever did for Waymo, because data is the only lever freight has.

Freight is Borrowing Credibility it Hasn’t Earned Yet

Now here’s where I have to be the guy who ruins the party, because this is a trucking blog and none of the numbers above are about trucks.

Every statistic in this piece so far comes from robotaxis. Passenger cars, in cities, at low speeds, with an enormous mileage base behind them. That IIHS study compared 50 million driverless miles against 222 billion human miles. Fifty million miles is a real sample. You can run statistics on it and trust the confidence rating.

Autonomous freight isn’t there. The most transparent trucking autonomy company on the road today is sitting somewhere around 440,000 cumulative driverless miles as of summer 2026, with a clean safety record over that distance. See Figure 3.

A clean record over 440,000 miles is genuinely good news. It is not the same thing as a statistically verified safety case. You need a lot more miles before “zero incidents” tells you anything more than “we haven’t gotten unlucky yet.”

Figure 3. The sample-size gap: passenger AV vs. freight AV

And trucks aren’t cars. A rig doesn’t stop like a sedan. It doesn’t have the same blind spots, the same rollover risk, or the same physics when it’s the one doing the striking instead of the one getting struck. Borrowing Waymo’s safety case to vouch for a truck is like citing a fighter jet’s flight record to clear a cargo plane for takeoff. They’re both aircraft. They fly in the same sky. But the payload, the handling, and the way each one fails when something goes wrong have almost nothing in common. Nobody signs off on a cargo plane using a fighter pilot’s logbook, and nobody should sign off on freight autonomy using a robotaxi’s.

If freight autonomy wants the public’s trust the way Waymo has started to earn it, it needs its own IIHS moment. Not a company press release. An outside group, checking the math, publishing the confidence intervals, telling us where the data is solid and where it still needs more miles.

I’m In, with Conditions

Let’s be honest about what the real goal is. Zero. Zero accidents, zero serious injuries, zero deaths on the road. That’s the only finish line worth naming out loud, and nobody should pretend a two-thirds reduction is good enough to stop chasing it. Every crash that still happens, even at a fraction of today’s rate, is still someone’s bad day or even someone’s worst day.

But here’s what I refuse to do, and what I don’t think this industry should do either: wait for perfect before we’ll accept better. If a technology can take a meaningful bite out of the crashes happening on our roads right now, today, while human drivers keep speeding and texting and driving tired, then sitting on our hands until it’s flawless has a cost too. That cost gets paid in real crashes, on real roads, by real people, every single day we wait.

I’m in. Not because the industry convinced me with a slick campaign, and not because I think the technology is finished. I’m in because the independent data, the kind that comes from IIHS and not from a company’s own marketing team, shows a real and measurable safety gain. That’s worth pursuing hard.

But my support comes with conditions, and they’re not negotiable. Keep the data honest. Keep testing rigorous. Keep publishing the confidence intervals right alongside the wins. And when something goes wrong — and something eventually will — be as loud about that as you were about the good news. That’s not just good ethics. It’s the only real insurance the industry has for the day the data isn’t the story anymore and one bad accident is. Trucking hasn’t earned its own version of that trust yet. The mileage isn’t there. The independent studies aren’t there. Given what’s coming down the road on self-certification and how these systems get validated before they’re trusted with 80,000 pounds at highway speed, that gap won’t close itself. It’s going to take the same thing that’s starting to close the gap for passenger AVs: data, testing, and transparency, in that order, with nothing skipped.