The Robotaxi Race Runs on Exemptions, Not Safety Standards

In the past six months, two companies received federal permission to charge passengers for robotaxi rides. One, Waymo, has 68% fewer police-reportable crashes per mile than human drivers across 50 million miles of operation ^[1]. The other, Tesla, crashed roughly once every 55,000 miles in Austin — about nine times the human rate ^[2]. Both were approved through the same mechanism: a case-by-case exemption from existing federal vehicle safety standards, with no performance threshold distinguishing the safer operator from the more dangerous one. That mechanism is not an exception. It is the model. Across the United States and Europe, regulators are not writing new operational safety rules for autonomous vehicles. They are waiving old hardware requirements — steering wheels, brake pedals, manual controls — and calling the waiver itself the deployment pathway. When NHTSA granted Zoox permission to operate a purpose-built robotaxi with no steering wheel or pedals this week, Transportation Secretary Sean Duffy framed it in precisely those terms.

This marks a major milestone towards providing the American AV industry with a streamlined pathway to scaled commercial deployment of novel AV fleets — Sean Duffy

The same word appears in Tesla's description of its path to European approval. The company told investors it was seeking "an exemption for the feature" through the Dutch vehicle authority RDW ^[3].

Our main path to success is partnering with the Dutch approval authority RDW to gain an exemption for the feature — Tesla

The Netherlands granted that type approval in April after 18 months of testing and 1.6 million kilometers of driving ^[3]. In June, the U.S. Department of Transportation proposed eliminating brake-pedal mandates for purpose-built autonomous vehicles, explicitly to "remove regulatory barriers" for Tesla's Cybercab and Zoox ^[4]. The same month, DOT withdrew a voluntary national evaluation framework that safety advocates had wanted for oversight ^[4]. The pattern is consistent: remove old hardware rules, do not replace them with new operational safety standards, and call the resulting gap the pathway to deployment. What fills that gap is industry self-regulation. Waymo built its own "Reference Driver" behavioral model with TU Delft, publishing it in Nature Communications, to benchmark AV safety against competent human driving ^[5]. It is the only company that voluntarily shared crash data with the Insurance Institute for Highway Safety for the study that produced the 68% figure ^[1]. IIHS President David Harkey warned of the limits of that arrangement.

However, the present data collection system isn’t good enough to allow continuous monitoring of a large-scale expansion. — David Harkey

The safety record, good or catastrophic, does not appear to change the regulatory mechanism. Waymo's robotaxis have blocked ambulances at mass-casualty scenes, illegally passed school buses 25 times in one Texas district, and been directed by a remote human agent to commit the same violation after a software recall meant to fix it ^[6][7]. The company has issued six recalls, including one for 3,871 vehicles that repeatedly entered closed freeway construction zones ^[8]. It has suspended services in six cities after vehicles failed to handle flooded roads ^[9]. And it continues to expand: San Diego, Las Vegas, Denver, Tampa, and 11 markets total ^[10]. NHTSA administrator Jonathan Morrison issued a formal warning that "an AV that cannot safely interact with first responders is a danger to the general public" — even as the agency continued processing commercial exemptions ^[10]. Tesla's trajectory is starker. Its Austin fleet logged 14 crashes in its initial phase, forcing a temporary pause in unsupervised operations ^[11][2]. NHTSA requested detailed fleet and safety data by June 19, citing concerns about crashes during reduced visibility ^[12]. U.S. Senators demanded an investigation into "weak and misleading" safety claims ^[13]. Former Tesla data labelers reported observing FSD-equipped vehicles speeding and failing to brake before striking animals or nearly hitting children; seven of nine said they would not ride in an FSD vehicle ^[14]. And yet Tesla expanded unsupervised service across the entire Austin metro area in June and now operates in seven U.S. markets, with autonomous miles growing over 10% weekly ^[12][15]. The exemption model has produced a bifurcated regulatory map. On one side, fast-track jurisdictions approve commercial operations through waivers: NHTSA, DOT, Texas, Nevada, and the Netherlands. On the other, resistance jurisdictions impose human-operator requirements or halt operations entirely. New York City let Waymo's permits expire ^[16]. Boston's city council proposed an ordinance requiring human safety operators in all autonomous ride-share vehicles ^[17]. New York State Senator Luis Sepulveda proposed banning driverless for-hire vehicles to protect over 100,000 drivers ^[18]. China suspended all new autonomous driving permits after more than 100 Baidu Apollo Go robotaxis froze on Wuhan streets, trapping passengers for up to two hours ^[19]. But the resistance jurisdictions are the exception. The prevailing model is the exemption. Between the two poles, a compromise tool has emerged: the supervised license. Transport for London granted private hire vehicle licenses to Wayve and Uber this week, allowing a robotaxi service with a human safety driver required in the driver's seat ^[20]. Uber launched its Las Vegas robotaxi service in March using Motional vehicles with human safety operators, expecting a transition to fully driverless by late 2026 ^[21]. Tesla's Austin launch used a mixed fleet — "a few unsupervised vehicles mixed in with the broader robotaxi fleet with safety monitors" — with social media footage showing human monitors following driverless cars in trailing chase vehicles ^[22]. The supervised license lets companies charge for rides without proving full autonomy. It is the regulatory expression of Uber CEO Dara Khosrowshahi's framing of the moment: a shift "from proving the technology to commercializing it at scale" ^[23]. The result is that the commercial robotaxi race is, at its core, a regulatory race. The same exemption model that lets Waymo expand to 11 markets lets Tesla expand to seven. The safety record — 68% fewer crashes than humans or nine times the human rate — does not change which mechanism applies. Proving the technology is not the gate. Securing the exemption is. And the exemption, once granted, has proven remarkably durable: recalls, warnings, probes, and suspensions have slowed deployment but not stopped it. The question the pattern raises is not whether any particular robotaxi is safe enough. It is whether a regulatory system built on waiving old hardware rules can ever answer that question at all.