The Question Nobody’s Answering Straight

Ask a trucking executive, a state trooper, or a member of Congress whether a fully autonomous truck can legally drive on American roads today, and you’ll get three different answers, all delivered with confidence. That’s the problem in one sentence.

“Legal” sounds like a yes-or-no question. It isn’t. Autonomous vehicle law in the United States is layered across three overlapping systems that were never designed to talk to each other: the safety rules governing how trucking companies operate (the Federal Motor Carrier Safety Regulations, or FMCSRs), the rules governing how vehicles are built (the Federal Motor Vehicle Safety Standards, or FMVSSs), and a patchwork of state traffic codes that predate the idea of a car driving itself by a century. Layered on top of all three sits an international treaty framework most Americans have never heard of, written in 1949 and 1968, that the United States is bound by whether or not anyone in Washington remembers it exists.

This brief gives the most direct answer the current state of the law allows: autonomous operation is legal in the narrow sense that nothing categorically bans it in most contexts, flatly illegal in at least one specific and important context, undefined at the level that matters most, and effectively unenforceable regardless of how the definitions eventually shake out. Each of those four claims gets its own section below, with the primary source behind it.

What Federal Trucking Law Says

The most thorough government answer to this question comes from a 2018 study by the Volpe National Transportation Systems Center, a research arm of the U.S. Department of Transportation. Volpe read through the entire FMCSR looking for language that would explicitly block a driverless truck. It didn’t find any.

But it found something almost as important: requirements that don’t explicitly demand a human driver, but implicitly assume one. Cargo re-inspection rules under 49 CFR 392.9 require a driver to physically check load securement at intervals during a trip. There’s no way to satisfy that requirement if nobody’s in the cab. Rules on vehicle inspection (Part 396) and vehicle condition (Part 393) run into the same wall. Volpe’s own words are worth repeating: the report found the agency “did not encounter specific language that explicitly requires a CMV to have a human driver present,” but that “other requirements… may implicitly require a human driver to be present.” A commercial motor vehicle, or CMV, is the regulatory term for the trucks and buses this whole framework governs.

Volpe hedged its own findings twice, warning that including driverless concepts in the analysis “does not suggest that they are likely, or even possible, under the current regulatory environment.” That’s a federal research center telling its own agency, in writing, not to read too much into the gaps it found.

The One Wall Nobody Can Waive Around

Buried in Part 397 of the FMCSRs is a requirement that turns out to be the single hardest legal barrier to driverless hazmat trucking anywhere in federal law, and it doesn’t get nearly enough attention.

Two separate provisions, 49 CFR 381.200(d) and 381.300(c), govern FMCSA’s authority to grant waivers and exemptions from its own rules. Both say relief “may only be granted from” a specific list of enumerated parts of the regulations. Part 397, which governs the transport of hazardous materials, is not on that list. Section 381.105 spells out exactly which parts count as the core safety rules for waiver purposes: 380, 382, 383, 384, plus pieces of 390, and Parts 391 through 393, 395, 396, and 399. Part 397 is conspicuously absent.

What that means in plain terms: FMCSA cannot use its normal administrative tools, the ones it uses for almost everything else, to grant a driverless hazmat carrier permission to operate. The only two routes around Part 397 are a formal rulemaking or an act of Congress. Until one of those happens, a fully autonomous truck cannot legally haul hazardous materials in the United States, and that’s not a gray area open to interpretation. It’s a closed door.

Both routes are in motion. In December 2025, the Pipeline and Hazardous Materials Safety Administration, or PHMSA, the federal agency that writes the hazmat rules, opened a formal process to revise those rules specifically for automated transport. Public comments closed in March 2026, and the agency hasn’t issued a final rule. Separately, a bill called the SELF DRIVE Act was introduced in the House in February 2026 that would create a federal framework for self-driving vehicles generally. It cleared a subcommittee vote and is sitting with the full House Energy and Commerce Committee. It hasn’t passed the House, hasn’t passed the Senate, and isn’t law. Similar bills failed in 2017 and 2021. So the door stays closed for now, but for the first time, both keys that could open it are being turned.

A Common Mistake Worth Correcting

A lot of writing on this topic treats Part 390, the FMCSRs’ definitions section, as a second barrier alongside Part 397. It isn’t, and the distinction matters.

Definitions don’t impose requirements. Section 390.5 defines “driver” as any person who operates a commercial motor vehicle, and defines “person” broadly enough to include a corporation. An automated driving system, or ADS, isn’t a person under that definition and can’t be a driver. But that’s a limitation on what an ADS can be called, not a rule the ADS has to comply with. Section 390.3(a) makes clear the rules apply to commercial motor vehicles as vehicles, not only to whoever’s driving them. There’s nothing in Part 390 itself to grant a waiver from, because it doesn’t require anyone to do anything.

Can a Machine Legally Be the “Driver”?

In February 2016, the National Highway Traffic Safety Administration, or NHTSA, the federal agency that sets vehicle safety standards, answered a question Google had been asking for months: could the software controlling its self-driving car count as the vehicle’s legal “driver”?

NHTSA’s chief counsel, Paul Hemmersbaugh, said yes, in writing. “If no human occupant of the vehicle can actually drive the vehicle, it is more reasonable to identify the ‘driver’ as whatever (as opposed to whoever) is doing the driving,” the letter read. “In this instance, an item of motor vehicle equipment, the [self-driving system], is actually driving the vehicle.”

That’s a real federal agency, in an official document, calling a piece of software the legal driver of a car. It made headlines at the time, and it’s still cited as a landmark today.

What gets left out of most retellings is how narrow the ruling was. NHTSA granted the software-as-driver interpretation for a specific list of provisions, mostly definitional ones, like where a warning light needs to be visible. For the provisions that mattered, like how the brakes get applied or how a turn signal gets cancelled, NHTSA said the interpretation alone didn’t solve anything, because the agency had no way to test or verify compliance. Google was told to pursue formal rulemaking or file for individual exemptions instead. Six years later, in a 2022 rule updating its crash-safety standards for automated vehicles, NHTSA deliberately declined to touch the core definition of “driver” at all. The agency gave four reasons, and they’re worth knowing because they show how deliberate the non-decision was. First, changing something as basic as the word “driver” needed more research than this one rule could cover. Second, the rule didn’t require settling the question, so there was no need to force it. Third, that fight belongs in a future rulemaking, maybe one that defines automated driving systems directly. Fourth, leaving the definition alone matched what a separate research project inside the agency had already recommended.

So the honest answer is: yes, a federal agency has said a machine can be a driver, and also, that same agency has spent the years since carefully avoiding making that answer permanent.

A Smaller Example, Same Pattern

A narrower but telling example: hazard flashers. Federal Motor Vehicle Safety Standard 108 requires a hazard warning system, and for decades the underlying industry standard defined it as “driver controlled.” NHTSA reversed course in a 2016 letter to General Motors, permitting automatic activation of hazard flashers without a human touching a button. That sounds like progress, and it is, but the permission only applies to Level 2 automation, where a licensed human is still fully in charge of the vehicle. The equipment requirement survived. Only the question of who’s allowed to press the button changed. NHTSA has said publicly that its own past interpretations on this kind of question “have reached different conclusions” over time, which is agency-speak for admitting the rules aren’t settled.

The State That Never Changed Its Law

Every U.S. state has its own vehicle code, and most have quietly updated theirs to make room for automated driving in some form. New York hasn’t.

New York Vehicle and Traffic Law Section 1226 requires “at least one hand” on the steering wheel “at all times when the motor vehicle is in motion.” The text hasn’t changed since 1971. Four separate bills to repeal or amend it died in the legislature between 2015 and 2020. A 2025 bill, A3650, doesn’t touch Section 1226 at all. Instead it creates a brand new article of law running parallel to the old one, leaving the hands-on-the-wheel requirement standing exactly as written.

New York is the only state with a requirement this explicit. It’s a live example of what happens when a law written for a world with no alternative simply never gets revisited, and it shows that “most states allow it” and “every state allows it” are very different claims.

What Happens When Theory Meets the Highway

In December 2024, NHTSA denied a request from Aurora Innovation, a self-driving trucking company, for an exemption covering three specific vehicle safety standards. NHTSA’s denial called the request “broad,” said it was “not supported by the data presented,” and found it “lacks necessary monitoring controls.”

Aurora deployed its driverless trucks on Interstate 45 in Texas anyway, on April 27, 2025, five months before its narrowed exemption took effect on October 9, 2025. Trade press covering the deployment reported, a week in, that “no clarity has been provided on how the company is prepared to comply.” Even after the exemption took effect, it covered exactly three narrow provisions. Nothing for driver qualification rules, drug and alcohol testing, hours-of-service limits, or most of the daily vehicle inspection requirements.

This is the gap between what the law says on paper and what happens on the ground, documented in real time by a real company operating real trucks on a real interstate.

The Treaty Layer Almost Nobody Talks About

Here’s the part of this story that rarely makes it into American coverage of self-driving trucks: the United States is bound by an international treaty that predates the interstate highway system.

The 1949 Geneva Convention on Road Traffic requires, in Article 4, that every vehicle have a driver, and defines driver as a person. The United States ratified it in 1950, with no reservations. Legal scholarship, most thoroughly a widely cited 2014 analysis by Professor Bryant Walker Smith, concludes the Convention is probably self-executing, meaning it operates as binding U.S. law without Congress passing implementing legislation, and that its driver requirement is flexible enough to plausibly cover a legal entity like a corporation rather than strictly a human being. But there’s a hard limit even to that generous reading: Smith is explicit that self-executing doesn’t mean privately enforceable. In practice, the Convention shows up in court as a defense, not as grounds for a lawsuit.

One provision of the Geneva Convention matters specifically for freight and gets skipped in most general-audience coverage: Article 5 excludes commercial carriage from parts of the treaty’s coverage. That’s a real limitation on how far this treaty framework reaches into trucking specifically, separate from the general driver question.

A Parallel Treaty, A Different Ending

A companion treaty, the 1968 Vienna Convention on Road Traffic, which the United States never ratified but which governs most of Europe and beyond, took a different path entirely. Where the American debate has stayed stuck in agency interpretation letters and quiet non-decisions, the international one reached a resolution.

Starting around 2016, the delegations that manage the Vienna Convention argued the question out loud, on the record, for years. Sweden and Belgium jointly proposed amendment language that would let a vehicle system satisfy the driver requirement directly. Germany and Japan said outright that amendment was necessary. The United Kingdom argued the existing text already had enough flexibility, pointing out that neither treaty ever defined “driver” as strictly human, since “person” can mean a corporation as well as an individual. France went further in 2018, proposing specific new treaty language after its own Council of State ruled that fully automated vehicles wouldn’t comply with France’s obligations under Vienna as written.

That process produced an actual answer. A new provision, Article 34bis, accepted in 2022, states that the driver requirement is satisfied when a vehicle operates under an automated driving system in line with a country’s own domestic technical rules. More than twenty countries, including Germany and France, have accepted it. The parallel Geneva Convention amendment effort, covering the treaty the U.S. is party to, failed to clear its own higher approval threshold. The two treaties that once said nearly the same thing now say different things, and the United States is bound by the one that never got fixed.

How One Country Drew the Line

The clearest short explanation of where the human-driver assumption comes from is Dutch, not American.

In 2017, the Netherlands’ highest advisory body on legislation, the Council of State, reviewed a bill permitting remote-driver testing of self-driving cars. Its reasoning: “At the time the relevant treaties and the Road Traffic Act came into being, control of a motor vehicle was only possible by a person who sat in the vehicle. It is therefore logical that those regulations assume there is a driver who must at all times have the vehicle in his power, but also that it is not explicitly stipulated that the driver must be present in the vehicle.”

The old laws never said the driver has to be physically inside the car. They just assumed it, because at the time there was no other option. A remote human driver, one who maintains control and can intervene instantly, fits inside that assumption without needing a single word of the law changed. A vehicle with no human driver at all does not, and the Council said so explicitly: once there’s no person driving, “international treaties in the field of traffic will then also require adaptation.” That’s the cleanest statement anywhere of exactly where the legal line sits.

The Strongest Case for the Other Side

Not every serious legal scholar agrees a machine can be a driver, even loosely. The most rigorous opposing case comes from Dutch legal scholar Nynke Vellinga, whose 2019 and 2020 work walks through the definition of “driver” in both treaties provision by provision.

Her argument: the treaties don’t just use “driver” as a label. The word carries rights and duties, capacity requirements like holding a license and being physically fit, and legal consequences when those duties are broken. A self-driving system has none of that. It can’t hold a license. It has no physical or mental condition to be fit or unfit. Vellinga argues that because the Conventions build an entire legal structure around a driver who can be licensed, tested, and held responsible, a piece of equipment simply doesn’t fit the definition no matter how capable it is at the driving task itself. On this specific point, she directly takes on NHTSA’s 2016 letter, arguing that even if a technical regulation can call software a driver, treaty law can’t, because the treaty’s “driver” carries obligations a machine can’t hold.

Her proposed fix, borrowed from Dutch legal theory, is to hold the manufacturer legally responsible for what the self-driving system does, the same way an aircraft’s pilot-in-command bears responsibility for a flight even during the stretches an autopilot is flying it. Whether or not you buy her ultimate conclusion, her reasoning is careful and worth taking seriously, not dismissing as an obstacle to progress.

The Question Nobody’s Built an Answer To

Suppose every definitional question above gets resolved tomorrow. There’s still a problem nobody’s solved: how does a police officer pull over a truck with no one behind the wheel?

A 2025 academic study by Brady, Tranter, and Bennett looked at roadside stopping powers, the legal authority officers use to make a vehicle pull over, and found nearly all of them are written as commands directed at a driver. “Powers directed to drivers to stop vehicles are problematic when applied to automated vehicles where the automated driving system cannot at law be considered a driver,” the lead author put it. Their research found one working model anywhere in the world: the United Kingdom’s Automated Vehicles Act 2024, which lets an officer communicate directly with a vehicle’s equipment and treats that as legally equivalent to directing a human driver. No American state or federal law does anything like it yet.

What the Regulator Itself Doesn’t Know Yet

FMCSA isn’t sitting on finished answers it’s declining to publish. It doesn’t have them yet.

The agency’s own research arm published a study titled “Driver Limitations in ADS-Equipped CMVs: An Analysis of Research Gaps,” which found, in its own words, “little research effort being undertaken specifically related to ADS-equipped CMVs.” A separate FMCSA rulemaking effort, tracked under regulatory identifier RIN 2126-AC17, has been projected as coming for years and still hadn’t been published as of mid-2026. FMCSA also launched a major new data-collection effort in 2025, the Crash Causal Factors Program, specifically because the government’s existing crash-cause data is more than two decades old and was never built with automated vehicles in mind.

The Straight Answer

Is autonomous trucking legal in the United States? Here’s the honest version, in four parts.

• Legal in most contexts, in the narrow sense that no federal rule explicitly bans it, and several agencies have shown willingness to interpret existing definitions flexibly.
• Illegal in at least one specific context: hazardous materials transport, where Part 397’s exclusion from waiver authority is a hard legal wall, not a gray area.
• Undefined at the level that matters most, because the agency with authority to settle the core “driver” definition question has repeatedly chosen not to, on the record, for reasons it has explained but not resolved.
• Unenforceable in practice regardless of how the definitions land, because nobody, state or federal, has built the roadside enforcement authority a driverless vehicle requires.

The international comparison matters here too. Other countries faced the same question years ago, argued it out loud through their legislatures and treaty bodies, and reached actual conclusions, even where those conclusions differed from country to country. The United States has mostly let individual agencies answer narrow slices of the question through interpretation letters and quiet non-decisions, while the deployment keeps moving ahead of the paperwork. Aurora’s trucks did not wait for Part 397 to get fixed. They didn’t wait for the exemption to take effect, either. That’s not a hypothetical risk. That already happened, on a real interstate, in 2025.

Sources

Volpe National Transportation Systems Center / FMCSA, “Review of the FMCSRs for Automated Driving Systems” (2018), FMCSA-RRT-17-013 / DOT-VNTSC-FHWA-18-02.

49 CFR 381.105, 381.200(d), 381.300(c), 390.3(a), 390.5, 392.9, Part 393, Part 396, Part 397.

PHMSA, Advance Notice of Proposed Rulemaking, “Hazardous Materials: Modernizing Regulations To Facilitate Transportation of Hazardous Materials Using Highly Automated Transportation Systems,” Docket No. PHMSA-2024-0064 (December 4, 2025).

H.R. 7390, SELF DRIVE Act of 2026, 119th Congress, introduced February 5, 2026.

NHTSA, “Google — compiled response to 12 Nov 15 interp request — 4 Feb 16 final,” nhtsa.gov/interpretations.

NHTSA, Final Rule, “Occupant Protection for Vehicles With Automated Driving Systems,” 87 FR 18560, Docket No. NHTSA-2021-0003 (March 30, 2022).

New York Vehicle and Traffic Law Section 1226; New York Assembly Bill A3650 (2025).

NHTSA denial of Aurora Innovation exemption petition (December 2024); Aurora Innovation public deployment announcements, April and October 2025; Land Line Media coverage, May 2025.

Convention on Road Traffic, Geneva, 1949, United Nations Treaty Series.

Bryant Walker Smith, “Automated Vehicles Are Probably Legal in the United States,” 1 Texas A&M Law Review 411 (2014).

Convention on Road Traffic, Vienna, 1968, as amended, including Article 34bis (accepted 2022, in force 2023).

UNECE Working Party on Road Traffic Safety (WP.1), Informal Document No. 4, 73rd Session (September 2016).

Advies Afdeling advisering Raad van State en Nader rapport, Kamerstukken II 2017/18, 34838, nr. 4 (Netherlands).

Nynke E. Vellinga, “Automated driving and its challenges to international traffic law: which way to go?,” Law, Innovation and Technology (2019); Vellinga, “Legal Aspects of Automated Driving: On Drivers, Producers, and Public Authorities,” University of Groningen (2020).

Mark Brady, Kieran Tranter, and Belinda Bennett, “Automated Vehicles, the ‘Driver Dilemma,’ Stopping Powers, and Paradigms of Regulating Road Traffic,” Computer Law & Security Review, vol. 56 (2025); Automated Vehicles Act 2024 (UK), section 57.

FMCSA, “Driver Limitations in ADS-Equipped CMVs: An Analysis of Research Gaps”; FMCSA Crash Causal Factors Program, Docket FMCSA-2025-0065, 90 FR 42059 (August 2025); RIN 2126-AC17, Unified Agenda.