Self-Driving Cars: Convenient or Risky?

When this post first ran, “self-driving” meant a prototype with a safety driver and a research permit. Six years later, driverless taxis carry paying passengers in four American cities, roughly half the new cars on Colorado roads steer and brake for themselves on the highway, and the term covers two completely different technologies that get confused constantly.

That confusion matters after a crash. A car that was driving itself and a car that was helping a distracted human are governed by different rules, produce different evidence, and point at different defendants. This post updates the picture for 2026: what the automation levels mean, what the crash data supports, what Colorado law says, and who can be held responsible when an automated system is at the wheel. Our Denver car accident lawyers at Boesen Law handle claims where the vehicle itself is part of the dispute.

“Self-Driving” Describes Six Different Things

The industry uses the SAE International automation taxonomy, which NHTSA presents as six levels. The gap between Level 2 and Level 4 is the one that decides claims.

• Levels 0 to 2 keep a human legally and practically in charge. Adaptive cruise control, lane centering, and combined systems such as GM Super Cruise and Ford BlueCruise sit here. The vehicle handles steering and speed under narrow conditions, and the driver must stay alert and take over at any moment.
• Level 3 lets the system drive without constant supervision in specified conditions, but a human must be in the driver’s seat and ready to respond when the car asks. Vehicles with genuine Level 3 systems remain very rare in the United States.
• Levels 4 and 5 are automated driving systems in the true sense. Level 4 does all the driving with no human supervision inside its operating conditions, which is what Waymo’s robotaxis do. Level 5, capable of driving anywhere a human could, does not exist yet in any production vehicle.

The practical point for Colorado drivers: nothing you can buy at a dealership today drives itself. As NHTSA puts it, even the highest level of driving automation available to consumers requires the full engagement and undivided attention of the driver. Marketing names like Autopilot and Full Self-Driving describe Level 2 systems.

For answers to your questions, call:
(303) 999-9999

What the Crash Data Actually Shows

The honest answer in 2026 is that Level 4 robotaxis are performing well in the places they operate, and Level 2 consumer systems have not yet shown a safety benefit.

• The Insurance Institute for Highway Safety published research in July 2026 finding that Waymo’s driverless vehicles crashed 68 percent less often per mile than human drivers in San Francisco, Phoenix, Los Angeles, and Austin, limited to crashes a person would normally report to police. The margin varied sharply by city: 76 percent lower in Phoenix, 71 percent lower in Los Angeles, 35 percent lower in San Francisco, and 4 percent higher in Austin, where the sample was small.
• The same study found Waymo vehicles involved in 85 percent fewer single-vehicle crashes and 81 percent fewer injury crashes per mile traveled than human drivers.
• IIHS was careful about the limits of that finding. Waymo logged about 50 million driverless miles over the study period against roughly 222 billion human-driven miles in the same locations, and Waymo is the only operator that reports its mileage voluntarily. Without mileage from the other companies, nobody can calculate their crash rates at all.
• For Level 2 systems, IIHS research on partial driving automation found no crash-reduction advantage over vehicles from the same automakers equipped only with crash avoidance features. The safety gains that do show up in the data come from automatic emergency braking, lane departure prevention, and blind spot detection, not from the automation layered on top of them.
• NHTSA recorded 39,254 people killed in motor vehicle crashes in 2024. Automation is not yet moving that number in either direction at national scale.

Nobody is entitled to state, on current evidence, that self-driving cars are broadly safer than people. What the evidence supports is narrower and more useful: one operator, in four metropolitan areas, under conditions it selects, has a materially lower police-reportable crash rate than human drivers.

The Level 2 Problem Is a Human Attention Problem

The most dangerous configuration on the road right now is not a driverless car. It is a competent Level 2 system paired with a driver who has stopped paying attention.

The NTSB investigation into the 2018 Tempe crash that killed a pedestrian remains the clearest case study. The automated driving system detected the pedestrian 5.6 seconds before impact but never classified her correctly or predicted her path, and the system design excluded emergency braking, relying on the human operator instead. That operator had been looking at a phone in the center console. The NTSB found the probable cause to be her failure to monitor the driving environment because she was visually distracted, with contributing factors including inadequate oversight of vehicle operators and no adequate mechanism for addressing automation complacency.

That pattern repeats in the research. IIHS has found that people who regularly use Level 2 systems often hold a false sense of security about what the technology can do, and that over-trusting these systems can stop drivers from intervening even when they can see a hazard developing ahead. IIHS began publishing partial automation safeguard ratings in 2024 precisely because driver monitoring, escalating alerts, and fail-safe behavior vary enormously between manufacturers.

For a Colorado claim, this is the analytically important point. When a Level 2 vehicle crashes, the human in the driver’s seat was legally driving. The system’s behavior may still matter, but it does not displace the driver’s duty.

Colorado Has a Statute on Automated Driving Systems

Colorado addressed this in 2017, and the provision still governs. Under C.R.S. § 42-4-242, a person may use an automated driving system to drive a vehicle or control a vehicle function if the system is capable of complying with every state and federal law that applies to the function it is operating. If the system is not capable of that compliance, it may not be tested on Colorado roads without approval from the Colorado State Patrol and the Colorado Department of Transportation, and a violation is a class B traffic infraction that can lead to the vehicle being impounded or immobilized.

Two features of the statute deserve attention. First, it disapplies provisions that by their nature regulate a human driver, except for laws regulating the physical driving of a vehicle. Second, and most importantly for anyone injured, subsection (5) says that liability for a crash involving an automated driving system driving a vehicle that is not under human control is determined in accordance with applicable state law, federal law, or common law.

In other words, Colorado did not create a special immunity and did not create a special remedy. Automated vehicle crashes are litigated with ordinary negligence and product liability tools.

Who Can Be Held Liable After an Automated Vehicle Crash in Colorado

There is rarely one answer. A properly built claim identifies every party whose conduct or product contributed.

• The human driver. In any Level 0 through Level 3 vehicle, the person behind the wheel owed the same duty of care they always did. Using a driver-assistance system does not transfer that duty to the manufacturer.
• The vehicle or system manufacturer. Colorado’s product liability statutes at C.R.S. § 13-21-401 and following define who counts as a manufacturer, and the definition is broad enough to reach a company that designs a component or exercises significant control over how it is produced. Defective design, defective manufacture, and failure to warn are all available theories.
• The software or sensor supplier. Perception, prediction, and planning are often built by companies other than the automaker, and the definition of manufacturer can reach a supplier of a component part.
• The fleet operator. For a robotaxi, the operator makes the decisions about where the vehicle runs, what conditions it accepts, how remote assistance works, and when a fleet is updated. That is operational conduct, judged under ordinary negligence.
• A commercial employer. If the vehicle was being used for work, the employer’s liability and coverage come into the case the same way they would after any commercial vehicle crash.
• A government road authority, where signal timing, signage, or lane markings the automated system relied on were themselves defective.

Colorado also builds in a rebuttable presumption that works against injured people and has to be planned for. Under C.R.S. § 13-21-403, a product is rebuttably presumed not defective if it conformed to the state of the art at the time of sale or complied with an applicable federal or state code, standard, or regulation. The same section runs the presumption the other way when a product did not comply. Because federal motor vehicle safety standards were largely written for human-driven cars, arguments about which standards even applied to a given automated feature become central. A Denver product liability lawyer has to be prepared for that fight from the beginning.

The Evidence in These Cases Is Different, and It Expires

An automated vehicle crash generates far more data than a conventional one, and almost none of it belongs to you.

• Event data recorder and telematics downloads showing speed, braking, throttle, steering input, and whether the automation was engaged in the seconds before impact.
• Automation state logs, meaning what the system detected, how it classified the object, what it predicted, and what it commanded. This is the record that decides whether the system or the human failed.
• Driver monitoring records, including inward-facing camera data and hands-on-wheel or eyes-on-road detection, which either supports or destroys the argument that the human was supervising properly.
• Over-the-air update history, because the software that was driving may have been replaced before anyone inspects the vehicle.
• Federal crash reports. Since 2021, NHTSA has required manufacturers and operators to report certain crashes involving automated driving systems and Level 2 driver-assistance systems under a Standing General Order, and the USDOT amended that order in 2025 as part of a broader automated vehicle framework.
• Corridor and roadside video, which is overwritten in days rather than weeks.

Most of this sits on servers controlled by the defendant. A preservation demand in the first weeks is what keeps it from disappearing on an ordinary retention schedule, which is also why accident reconstruction in these cases starts with a data fight rather than a tape measure.

Deadlines Are Shorter Than Most People Expect

Colorado gives motor vehicle injury claims three years from the date of the crash under C.R.S. § 13-80-101. Claims against a manufacturer or seller of a product run on a different and shorter clock. C.R.S. § 13-80-106 requires actions against a manufacturer or seller for personal injury caused by a product to be brought within two years after the claim for relief arises, whatever legal theory is used.

An automated vehicle crash can easily involve both. Treating the shorter deadline as the operative one is the only safe approach, and the practical deadline is shorter still, because the data described above does not wait for a statute to run. Our overview of time limits on personal injury cases explains how the periods interact.

An Attorney’s View on Automated Vehicle Claims

Jon C. Boesen, founder of Boesen Law, has 36 years of civil litigation experience and has overseen mass tort and class action work involving defective drugs and products. He notes:

“These cases look futuristic and they are not. They are product cases with a car attached. The questions I ask are the same ones I have asked about a defective drug for three decades: what did the company know, when did it know it, what did it tell the user, and what did it choose not to fix. What is genuinely new is that the proof lives in a log file the defendant controls and can overwrite with an update. So the first move is never a demand letter. It is a preservation letter, and it goes out before anyone has decided what the case is worth.”

What Cases Like Yours Have Recovered

Automated systems do not change what a serious injury is worth, and Boesen Law’s motor vehicle results reflect the work that goes into proving one. Our case results include a $770,000 recovery for a client struck by a driver who ran a stop sign, a $525,000 recovery for a client who was T-boned by a red-light runner and hospitalized with concussion symptoms and lasting back and neck pain, and a $1,450,000 recovery for a client side-struck on Arapahoe Road by a driver traveling at 90 miles per hour.

Contact a Denver Car Accident Lawyer at Boesen Law

If an automated or partially automated vehicle was involved in your crash, the defendant already has the data and you do not. That imbalance closes only when someone sends the letters and asks the right questions early, while the logs, the update history, and the roadside video still exist.

Boesen Law is a boutique firm with big results, our attorneys and staff work in English, Spanish, and Russian, our phones are answered at any hour, and there is no attorney fee unless we recover compensation for you. Contact Boesen Law for a free, in-person consultation, and bring whatever you have, including photographs, the report number, the make and model of the other vehicle, and anything the other driver said about what the car was doing.

FAQs About Self-Driving Cars and Colorado Crash Claims

Are self-driving cars legal in Colorado?

Yes, within limits set by statute. C.R.S. § 42-4-242 permits the use of an automated driving system to drive a vehicle or control a vehicle function so long as the system can comply with every state and federal law applying to the function it is operating. A system that cannot meet that standard may not be tested on public roads without approval from the Colorado State Patrol and the Colorado Department of Transportation, and violating that requirement is a class B traffic infraction that can result in the vehicle being impounded. The statute also leaves crash liability to ordinary state, federal, and common law rather than creating a separate regime.

If the car was driving itself, is the person in the driver’s seat still responsible?

Usually, yes, because almost no consumer vehicle is actually driving itself. Systems marketed under names like Autopilot, Full Self-Driving, Super Cruise, and BlueCruise are Level 2, which means the human is the driver and must remain ready to take over at any moment. Colorado law does not excuse a driver who delegated attention to a system that was never designed to accept it. The only vehicles genuinely operating without a human driver are Level 4 fleet vehicles such as robotaxis, and those are not sold to the public.

How do I prove the automation failed rather than the driver?

Through the vehicle’s own records, not through argument. Event data recorder downloads, automation state logs, driver monitoring data, and over-the-air update history together show whether the system was engaged, what it detected, what it commanded, and whether the human responded. Corridor video and an independent reconstruction corroborate that record. The obstacle is access, since nearly all of it is held by the manufacturer or fleet operator on retention schedules that do not accommodate injured people. Sending a preservation demand early is what makes the rest possible.

Are driverless taxis actually safer than human drivers?

The best current evidence is encouraging and narrow. IIHS research published in July 2026 found that Waymo’s driverless vehicles were involved in 68 percent fewer police-reportable crashes per mile than human drivers across San Francisco, Phoenix, Los Angeles, and Austin, with 81 percent fewer injury crashes. The result covers one operator, four cities, roughly 50 million driverless miles, and conditions the operator chooses. IIHS itself cautioned that the national reporting system is not good enough to confirm that performance holds as these fleets expand, since most companies do not report the miles they drive.

Does my auto insurance cover a crash involving an automated vehicle?

Your own coverage responds the way it normally would, including medical payments coverage and uninsured or underinsured motorist benefits if the other party is inadequately insured. What changes is where the rest of the recovery comes from. A claim against a manufacturer, a software supplier, or a fleet operator is a product or negligence claim against a company, not an ordinary auto claim, and those defendants carry very different coverage. Identifying every available source before anyone discusses settlement is what keeps a serious injury from being measured against a single policy limit.

How long do I have to file a claim after an automated vehicle crash in Colorado?

It depends on who you are suing, and the safest assumption is the shorter period. Motor vehicle injury claims generally carry three years from the date of the crash under C.R.S. § 13-80-101. Claims against a manufacturer or seller of a product must be brought within two years after the claim for relief arises under C.R.S. § 13-80-106, regardless of the legal theory. Because an automated vehicle crash frequently supports both kinds of claim, working to the two-year deadline avoids losing the product case while preserving the other. The evidence timeline is far tighter than either.

Call (303) 999-9999 or complete a Free Case Evaluation form