During the week of November 17, 2026, American search interest in “waymo” jumped as national press reported that California’s Department of Motor Vehicles had barred Waymo robotaxis from the state’s freeways and that the company was pausing driverless freeway rides in every market where it operates. According to those press reports, the state action followed two freeway episodes: three Waymo vehicles entering a coned-off construction zone on Route 101 near Burlingame, and a Waymo vehicle attempting to drive around a California Highway Patrol traffic break on Interstate 280, an episode widely shared on video. No official restriction order was read during the preparation of this article, so the scope and duration of the freeway restriction below are reported as the press described them, not as established regulatory text. What is established, from documents read for this article, is the legal frame around the story: California permits driverless deployment only through its own approvals, federal safety law gives the national traffic-safety agency investigative and recall power over automated driving systems, and courts have begun to decide who answers when driving software, rather than a human driver, makes the mistake. The discussion that follows is general in scope: it describes checks useful in comparable situations and makes no findings about Waymo’s situation.
I. Can a driverless taxi legally use the freeway, and who gets to decide?
A. In California, deployment is a permission, not a right
California law treats an autonomous vehicle as a specific legal category. Section 38750 of the Vehicle Code defines “autonomous technology” as technology with the capability to drive without active physical control or monitoring by a human operator, and an “autonomous vehicle” as a vehicle carrying technology meeting Levels 3, 4, or 5 of the SAE J3016 taxonomy, while expressly excluding ordinary driver-assistance features such as adaptive cruise control or lane-keep assist that cannot drive the vehicle on their own. The same section provides that, outside testing, an autonomous vehicle shall not be operated on public roads until the manufacturer submits an application to the Department of Motor Vehicles and that application is approved under department regulations. Testing itself requires proof of insurance, a surety bond, or self-insurance of five million dollars. Read the definitions and the approval requirement at California Vehicle Code section 38750.
The statute also answers the question every crash case asks first: who counts as the driver when nobody is driving. An “operator” of an autonomous vehicle is the person seated in the driver’s seat or, if no one is in that seat, the person who causes the autonomous technology to engage, while a “manufacturer” of the technology is the company that originally equips it or modifies a vehicle with it. Those definitions matter because a fully driverless taxi has an operator that is a company, not a person behind the wheel, which pushes claims away from ordinary driver negligence and toward the doctrines that govern companies that put products on the road. The department may also demand additional certifications in the deployment application, including how the maker will handle a failure of the technology, and it writes the detailed testing and deployment regulations the application must satisfy.
That state permission sits beside a second California approval for passenger service. On August 14, 2026, the Consumer Protection and Enforcement Division of the California Public Utilities Commission approved Waymo LLC’s Advice Letter 0004 and its supplement, accepting an updated Passenger Safety Plan tied to an expanded operational design domain and a vehicle platform the Department of Motor Vehicles had approved, with the deployment permit amended on November 21, 2025. The letter, addressed to Waymo LLC of Mountain View, California, also records protests from San Diego transit bodies, supportive responses, and added procedures on unaccompanied minors and rider safety during operational disruptions. Read the disposition at the Commission’s August 14, 2026 disposition letter. A deployment authorization therefore describes where and how a driverless fleet may run, and the November 2026 press accounts describe California narrowing that authorization by removing freeways. Because the restriction text itself was not read here, a rider or driver sharing the road should treat press summaries as a signal to check the current permit status with the department rather than as the operative legal language.
California thus runs a two-key system: the motor-vehicle department licenses the testing and deployment of the vehicle itself, while the utilities commission authorizes carrying passengers for compensation, including the safety plan for an expanded operating domain. The operating domain is the practical heart of both approvals, because it lists the conditions the system is approved to handle: which cities, which road types, which speeds, which weather. A freeway exclusion redraws that boundary after the fact, and the press accounts describe exactly such a redrawing, with surface-street service continuing while freeway operation stops. For an injured person, the boundary matters twice: it determines whether the vehicle was operating inside its approval at the moment of the crash, and it shapes what the operator knew about the limits of its own system.
B. Washington watches the software: crash reports, investigations, and recalls
Federal law gives the Secretary of Transportation broad power over motor-vehicle safety. Title 49 of the United States Code defines the key terms of that system, directs the Secretary to reduce traffic deaths and injuries, and, in section 30166, authorizes the Secretary to require manufacturers to keep records and make reports so the department can decide whether the law is being obeyed. Read the definitional section at 49 U.S.C. section 30102 and the records-and-reports authority at 49 U.S.C. section 30166. One concrete use of that power matters directly for robotaxis: on April 5, 2023, the National Highway Traffic Safety Administration issued a Standing General Order requiring manufacturers to report crashes occurring when certain automated technology was in use shortly before or at the time of the crash, compiling redacted aggregates for its public website.
That reporting system was tested in court in Wp Company LLC v. National Highway Traffic Safety Administration, No. 24-cv-1353 (D.D.C. Mar. 25, 2026), decided by the United States District Court for the District of Columbia. The opinion, by Judge Tanya S. Chutkan, explains that under the National Traffic and Motor Vehicle Safety Act the agency is responsible for reducing traffic accidents and injuries, describes the April 5, 2023 Standing Order and its redacted public spreadsheet, and resolves cross-motions for summary judgment in a Freedom of Information Act fight over the underlying crash data, in which a carmaker had intervened: the court will GRANT in part and DENY in part Defendants’ Motions for Summary Judgment and DENY Plaintiff’s Cross-Motion for Summary Judgment without prejudice. For a reader, the useful application is direct: automated-driving crashes are already funneled into a federal reporting and oversight system, and the agency publishes only redacted aggregates, so the absence of a public spreadsheet row never proves a crash did not happen, and a published row never proves a defect on its own.
The data fight behind that oversight deserves attention, because it limits what the public can learn from any single crash. The newspaper sought unredacted crash reports, the agency withheld business-confidential and personal material, and the vehicle maker intervened to protect its own submissions, leaving the published spreadsheet as an aggregate with redactions. An investigation file moves in stages, from a preliminary evaluation such as PE25013 through deeper engineering analysis toward a recall or a closing, and each stage reflects the evidence gathered so far rather than a final verdict on the technology. The practical lesson is to read agency files for what stage they record: an opened evaluation proves scrutiny, a recall proves a defined defect finding, and neither proves how a later, different crash happened.
When reports suggest a safety problem, the agency can investigate and the manufacturer can be forced into a recall. The Office of Defects Investigation opened Preliminary Evaluation PE25013 on October 17, 2025, prompted by media reports, concerning a Waymo 5th Generation automated driving system in an estimated population of 2,000 vehicles, after a documented episode in which a Waymo vehicle passed a school bus in Atlanta, Georgia on September 22, 2025; the file recorded zero crashes, zero injuries, and zero fatalities at opening. Read the opening resume at NHTSA Preliminary Evaluation PE25013. Separately, Waymo LLC filed a Part 573 Safety Recall Report numbered 25E-034 on May 12, 2025, covering 1,212 vehicles with 5th Generation software predating a November 2024 update, after an earlier evaluation of reports that included collisions with gates, chains, and barriers. Read the filed report at NHTSA Recall Report 25E-034. The governing statute provides that once the Secretary initially decides a vehicle contains a safety-related defect, the Secretary notifies the manufacturer, and the federal recall regulation requires the manufacturer to identify the recall population and the basis for drawing its boundaries. Read the duty at 49 U.S.C. section 30118 and the reporting rule at 49 CFR Part 573. A recall, in other words, is a finding about a defined population of vehicles, not a verdict on every later crash.
The recall notice itself is a regulated document. The statute requires it to state that the manufacturer giving notice will remedy the defect or noncompliance without charge under section 30120 of this title, to give the earliest date the free remedy will be available, and to explain the procedure for telling the Secretary when a manufacturer fails to remedy. That structure explains why recall 25E-034 reads the way it does: it names the manufacturer, describes the equipment and the software vintage, states the size of the affected population, and chronicles the agency evaluation that prompted it. A driver checking whether a crash vehicle was covered therefore looks for those four elements, make and model year, software vintage, population boundaries, and date range, before drawing any conclusion about a different crash months later.
II. If a driverless car crashes on the highway, who pays?
A. The liability paths courts actually use: defect, warning, and proof of knowledge
Start with the hardest lesson from heavy-vehicle crash litigation, which transfers directly to a driverless highway crash: blaming the machine is not enough; the plaintiff must prove a defect and, for the heaviest penalties, the maker’s knowledge. In Leslie versus Daimler Trucks North America LLC, No. 24-11372 (11th Cir. July 15, 2025), decided by the United States Court of Appeals for the Eleventh Circuit, a driver who crashed a heavy truck into stopped traffic at freeway speed and suffered burns over most of his body sued the manufacturer for failure to warn and negligent fuel-tank design. The court affirmed summary judgment for the manufacturer on both claims, holding that under Georgia law there was no duty to warn of the open and obvious risk that a freeway-speed crash could cause a fire, and that failing to adopt theoretical safety mechanisms no other heavy truck used could not show the reckless disregard for life that punitive damages require. The application for a driverless highway crash is that courts measure warnings against what an ordinary road user would already appreciate and designs against what the industry actually uses: a danger obvious to any freeway driver needs no warning, and an untested safety theory, without more, does not prove the knowledge that unlocks the heaviest penalties.
Punitive damages sit at the top of that ladder, and the Eleventh Circuit shows how high the rung is. The opinion explains that theoretical evidence about safety mechanisms the maker could have adopted, without proof the maker knew inaction would very probably injure someone, leaves a jury able to conclude at most that prevention was possible, which cannot by itself prove reckless disregard. The court treated industry practice as the counterweight: mechanisms tested or adopted by no other heavy truck in the country could not establish the required knowledge. For a highway claimant, the message is symmetrical: gather the prior-incident record and the internal knowledge evidence early, because without them the case stays in compensatory territory, while documenting warnings the rider actually received matters just as much for valuing what the claim can realistically carry.
California’s highest court sharpened the same point for design claims generally in Gilead Tenofovir Cases, No. S283862 (Cal. Aug. 3, 2026), decided by the Supreme Court of California. The court reversed the Court of Appeal and remanded with directions to enter summary judgment for the manufacturer, holding that a maker does not owe a duty of care to users of a nondefective product when deciding whether and when to commercialize an allegedly safer alternative, because that decision is remote from a plaintiff’s injuries caused by an existing drug, while negligence still requires proof that the maker’s fault produced a defective product and warning duties attach to known or reasonably knowable risks. The application is that a safer software update existing in theory does not itself prove the shipped version defective: a claimant must still prove the version on the road was defective when it crashed, or that the maker failed to warn of a risk it knew or should reasonably have known, measured against the state product-liability law of the crash state.
The California decision adds the duty framework that negligence claims must survive. Starting from the default rule that everyone must use ordinary care to prevent foreseeable harm, the court explained that foreseeability and policy considerations may justify an exception to section 1714’s default duty in appropriate circumstances, and found such circumstances where the allegedly safer product had never completed large-scale human testing or received federal approval. It left the ordinary paths intact: negligent design or manufacture proven against a product shown to be defective, and failure to warn of risks the maker knew or could reasonably have known. Translated to driving software, the question is never whether a better version was imaginable in hindsight; it is whether the version on the road was defective when measured against feasible alternatives and known risks at the time it drove.
Three practical consequences follow for a driverless highway crash. First, with no human in the driver’s seat, there is no driver negligence to absorb the loss; claims run against the entities that put the vehicle on the road, principally the manufacturer-operator, under the product-liability and negligence law of the state where the crash occurred, with California’s definitions of operator and manufacturer in section 38750 as the starting point for who counts as what. Second, warnings and instructions matter on both sides: documented rider-facing warnings and a complied-with safety plan help the operator, while proof the system could not handle a foreseeable highway event, such as a coned construction zone or an emergency traffic break, helps the claimant. Third, a regulator’s restriction is evidence of concern, not a finding of defect or causation; no court in the decisions read here treats an agency limit as a substitute for proving what failed in a particular crash.
B. The checks that protect a claim in the first days after a crash
The first hours decide what a lawyer can prove months later, so work in this order. Get medical care immediately and call the police so an official report exists; highway collisions with injuries also draw the patrol agency whose traffic-break or construction-zone orders may become central evidence. Photograph everything before vehicles move: lane markings, cones and signs, skid marks, damage to every vehicle, and the robotaxi’s exterior and interior screens. Keep the ride receipt, the trip record in the application, and any message the operator sent about the trip; write down the time, mile marker, and direction of travel while memory is fresh. Do not give the operator’s insurer a recorded statement before understanding the coverage picture, and do not sign a release at the roadside.
Next, preserve the data only the operator holds. Automated vehicles log sensor, mapping, and disengagement data that can be overwritten in routine operation, so a prompt written preservation request, followed by counsel’s formal litigation-hold letter, asks for the trip logs, software version, remote-assistance records, communications with emergency responders, and any disengagement or minimal-risk-condition maneuver the vehicle executed. File a vehicle complaint with the national traffic-safety agency, which feeds the same reporting system described above, and keep a copy of the submission. Track down independent witnesses and any patrol-dash or highway-camera footage quickly, because public video is routinely purged within weeks.
Finally, map the money before the bills compound. Identify every potentially responsible party: the manufacturer-operator that deployed the vehicle, any company that modified or maintained the driving system, a construction contractor whose zone signing may have confused the system, and, where another vehicle contributed, that driver. Match each candidate to its coverage: the operator’s commercial policy, the testing-or-deployment security California law demands, your own uninsured and underinsured motorist coverage, and health coverage for immediate bills. Time limits for injury claims are short and differ by state and by type of claim, and missing one can end a case regardless of merit, so confirm the controlling deadline for the crash state at once rather than assuming the general rule travels. Keep every document, photograph, receipt, and message in one file, because the gap between a barred freeway and a paid claim is filled entirely with proof.
Separate the property claim from the injury claim from the start, because they travel on different clocks and different coverage. Photograph the damage, get repair estimates from shops rather than verbal guesses, and keep rental receipts, because transportation costs during repair are recoverable losses in many states when documented. For injuries, keep every emergency-room record, follow-up appointment, prescription, and physical-therapy order, plus pay stubs or employer letters proving missed work; pain and limitation journals written day by day carry more weight months later than memory alone. Bring that file, the police report number, the ride receipt, the agency complaint copy, and the preservation letters to the first serious conversation about the case, and ask directly which deadline controls each claim, which coverages have been notified, and what additional evidence, such as corridor camera footage or patrol video, still needs a formal request before it is purged.
Conclusion
A freeway restriction answers where a driverless fleet may run; it never answers who pays for a particular crash. That answer still comes from state product-liability and negligence law applied to preserved facts: whether the driving system was defective, what its operator knew and warned about, and which insurers stand behind the vehicle, the roadway work, and the injured person’s own policy. Report the crash, protect the data, and pin down the deadline early, and the legal question becomes provable instead of merely debatable.