Autonomous vehicles no longer fit neatly into the categories of research prototype or conventional car. Driverless passenger services operate in selected cities, while testing expands to new roads and weather conditions. That progress does not mean the technical problem is solved. It does mean the decisive questions increasingly sit outside the vehicle: who decides that a system is ready, which evidence counts, how incidents are investigated, and what authority a city has over machines using its streets.

Capability is conditional, not universal

An automated driving system operates within an operational design domain: a defined combination of geography, road type, speed, weather, and other conditions. Success inside that domain is not proof that the vehicle can drive everywhere. California reported more than nine million autonomous test miles on public roads between December 2024 and November 2025, but its figures exclude private-road, out-of-state, simulation, and automation below SAE Level 3 (California DMV). Even a large mileage total therefore describes a bounded program, not general driving intelligence.

Commercial evidence is becoming more substantial. Waymo reported 271.3 million fully autonomous miles through June 2026 and said its vehicles had 82% fewer injury-causing crashes than a matched human benchmark across five metropolitan areas (Waymo). Those are encouraging company-reported results, but the comparison depends on exposure, geography, crash definitions, reporting completeness, and benchmark construction. Governance must make such methods inspectable and comparable across operators.

No single regulator owns the whole system

In the United States, federal authorities regulate vehicle safety and can investigate defects. States license vehicles and authorize testing or deployment. Cities manage streets, curbs, emergency response, construction, and transit interactions. Public-utility commissions may regulate paid passenger service. A driverless fleet touches all of those responsibilities at once.

California illustrates the layered model. The Department of Motor Vehicles issues testing and deployment permits, while the Public Utilities Commission separately authorizes passenger programs and requires driverless applicants to submit passenger safety plans (CPUC). This division can provide multiple checks, but it can also leave local officials responsible for blocked roads or emergency scenes without direct control over fleet deployment.

At the federal level, the National Highway Traffic Safety Administration requires named manufacturers and operators to report certain crashes involving automated driving systems. Its order was amended in 2025, and the agency warns that reports can contain duplicates, unknowns, and unequal operator access to incident information (NHTSA). NHTSA's public explanation contains a revealing sentence: “Vehicles are tested by the companies that build them” (NHTSA). A credible regime cannot depend only on a developer's private readiness decision.

The safety case must survive an incident

Predeployment simulation, closed-course tests, safety cases, and scenario libraries are essential, but public legitimacy is formed after failures. In 2024, NHTSA imposed a $1.5 million penalty on Cruise after finding incomplete reports about an October 2023 crash in which a driverless vehicle dragged a pedestrian approximately 20 feet (NHTSA). The episode showed that accurate disclosure is part of safety, not a separate communications issue.

Autonomous Vehicles Face a Harder Governance Test Than a Technical One — supporting image
Autonomous Vehicles Face a Harder Governance Test Than a Technical One — supporting image

New events keep testing the system. The National Transportation Safety Board opened investigations after a Waymo vehicle passed a school bus loading students in Austin on January 12, 2026, and after another struck a child in a Santa Monica school zone on January 23 (NTSB). An investigation is not a finding of fault. It is evidence that unusual, high-consequence interactions require an institution capable of reconstructing what happened and issuing enforceable lessons.

The NTSB has long recommended safety-management systems, stronger testing applications, and federal evaluation of developer safeguards. It also emphasizes that humans are poor monitors of partial automation, a separate but related problem when marketing blurs the line between assistance and autonomy (NTSB). Clear consumer language is therefore a governance control.

Public rules should scale with deployment

A mature framework would require a documented safety case before service, standardized exposure and incident data, prompt preservation of logs, independent access for investigators, cybersecurity and update controls, accessible passenger support, and plans for emergency responders. Material expansion into a new geography or vehicle platform should trigger review rather than being treated as a routine software release.

Local performance matters too. A vehicle can avoid collisions while still obstructing a fire scene, stopping in an unsafe place, or overloading a curb. Cities need structured channels to report recurring behavior and receive timely remedies. Operators need predictable rules rather than a patchwork of ad hoc demands. The solution is not to assign every technical decision to city hall, but to connect national safety oversight with local operational knowledge.

Autonomous vehicles may ultimately reduce road harm, and growing real-world evidence deserves serious attention. The hardest test is whether institutions can verify that benefit while systems change rapidly and deployments expand. A technically capable vehicle earns permission to operate only when its developer is accountable, its evidence is legible, and the public retains meaningful authority over the roads it enters.

Sources