A Robotaxi Swept Into a Texas Creek Triggered a 3,791-Vehicle Recall — and a Reckoning for Autonomous Vehicle Regulation

On April 20, 2026, a Waymo robotaxi traveling a 40 mph highway in San Antonio, Texas, encountered a flooded stretch of road. The vehicle's sensors detected the water. Its software commanded a deceleration. Then it drove in anyway. The empty car was swept off the roadway and into Salado Creek, where it sat submerged for days before crews could retrieve it [1][2].

No passengers were aboard. No one was injured. But the consequences have rippled through the autonomous vehicle industry. Waymo has recalled 3,791 robotaxis across its entire U.S. fleet, paused operations in San Antonio for its longest service stoppage to date, and handed federal regulators a case study in how a single edge-case failure can reshape the regulatory landscape for self-driving cars [3][4].

What Happened in San Antonio

The April 20 incident was not Waymo's first encounter with San Antonio's flash-flood-prone roads. Approximately two weeks earlier, another unoccupied Waymo vehicle became stranded in floodwaters near McCullough Avenue and Contour Drive [2]. That vehicle was recovered without being swept away, but the back-to-back events pointed to a systemic vulnerability rather than a one-off sensor glitch.

In its filing with the National Highway Traffic Safety Administration (NHTSA), Waymo described the root cause as a software condition specific to higher-speed roadways: when the automated driving system encountered standing water it classified as untraversable, it would slow down but continue forward rather than stopping or rerouting entirely [3][5]. The vehicle detected the hazard. The gap was between perception and action — the system recognized danger but executed an inadequate response.

Video footage from Austin and other cities had previously documented Waymo vehicles entering flooded streets or stalling during heavy rainstorms, suggesting that the flood-response deficiency was not unique to San Antonio's geography [6].

The Recall: Scope and Mechanics

Waymo filed a voluntary recall with NHTSA on April 30, 2026 — ten days after the Salado Creek incident [2]. The recall covers all 3,791 vehicles in Waymo's commercial fleet equipped with fifth- and sixth-generation automated driving systems, spanning every city where the company operates: Phoenix, San Francisco, Los Angeles, Austin, San Antonio, and Atlanta [3][7].

Unlike traditional automotive recalls that require vehicle owners to visit a dealership, Waymo's fix is delivered as an over-the-air (OTA) software update — comparable to a smartphone security patch [7]. Waymo stated that on the same day as the April 20 incident, it deployed interim operational controls including new weather-related constraints and revised mapping data [6]. A "final remedy" with additional software safeguards is still under development [5].

The company paused all San Antonio operations following the incident — a suspension that lasted roughly three weeks and represented the longest service interruption in any Waymo market to date [8][9]. Waymo announced plans to resume San Antonio service around May 10, after completing a review of its flood-monitoring procedures [8].

Waymo has not disclosed the direct cost of the recall or the revenue lost during the San Antonio pause. Because the fix is software-based and the fleet is company-owned, the per-vehicle remediation expense is minimal compared to a traditional mechanical recall. The greater financial exposure likely comes from reputational damage and the potential for regulatory friction in new markets [6].

Waymo's Safety Record in Context

The flood recall is Waymo's third voluntary safety action in two years. In February 2024, the company recalled approximately 1,200 vehicles after a software defect caused collisions with parked cars and stationary objects. In May 2024, another recall covering 670 vehicles followed a low-speed collision with a telephone pole in Phoenix, attributed to faulty mapping data [10].

Source: NHTSA recall filings

Data as of May 13, 2026CSV

These numbers are small compared to the autonomous vehicle industry's largest recall: Tesla's December 2023 action covering more than 2.4 million vehicles over Autosteer misuse concerns, prompted by nearly 1,000 reported crashes [10]. Amazon's Zoox recalled 270 vehicles in April 2025 after a crash in Las Vegas [10].

Yet Waymo's overall safety statistics remain strong. A peer-reviewed analysis of 7.1 million rider-only miles found an 85% reduction in injury-involved crashes and a 57% drop in police-reported crashes compared to human drivers in the same operating zones [11]. A subsequent Swiss Re study covering 25 million miles reported 92% fewer bodily injury claims and 88% fewer property damage claims [12]. In Phoenix specifically, Waymo's injury crash rate stood at 0.6 per million miles versus 4.9 for human drivers [12].

Source: Waymo Safety Impact Report / Swiss Re

Data as of Jan 1, 2026CSV

Over 127 million fully autonomous miles driven through September 2025, Waymo has accumulated more than 150 human driving lifetimes of road experience [12]. Its airbag deployment rate is 23 times lower than that of human drivers per million miles [12].

The tension at the heart of this recall is whether a single dramatic failure — a car swept into a creek — should weigh more heavily in public perception and regulatory action than a statistical safety record that outperforms human drivers by wide margins.

The Regulatory Framework: What Authority Did NHTSA Use?

Waymo classified this as a voluntary recall, meaning the company initiated the NHTSA filing rather than being compelled by a formal defect order [3]. This distinction matters. Under the National Traffic and Motor Vehicle Safety Act, NHTSA can compel a recall if it determines a vehicle contains a safety-related defect or fails to comply with Federal Motor Vehicle Safety Standards (FMVSS). But NHTSA also retains full defect, recall, and enforcement authority over autonomous vehicles — a power the agency has affirmed repeatedly as AV deployment scales [13][14].

The federal framework for autonomous vehicles is in active flux. In April 2025, Transportation Secretary Sean Duffy introduced a new AV Framework prioritizing safety, innovation, and commercial deployment [14]. NHTSA has announced plans for four proposed rulemakings in 2026 to reduce design barriers for vehicles without manual controls [14]. Meanwhile, the SELF DRIVE Act of 2026 (H.R. 7390), introduced in the House of Representatives, would direct NHTSA to establish updated safety standards specifically for automated driving systems, require manufacturers to submit detailed "safety cases," and preempt conflicting state manufacturing bans [13].

The bill would not, however, create new liability rules or shield compliant manufacturers from common-law lawsuits [15]. This gap leaves the liability question for incidents like the Salado Creek flooding squarely in the domain of existing product-liability doctrine — a framework designed for hardware defects, not software decision-making.

NHTSA's current crash-reporting Standing General Order (SGO), which requires AV companies to report certain incidents, is set to expire in 2026. Congress may consider directing NHTSA to extend it [13]. If the SGO lapses, the public's ability to track autonomous vehicle safety data would diminish at precisely the moment when fleet sizes are growing.

Who Was Harmed, and What Legal Recourse Exists?

In the San Antonio incidents, no passengers were aboard and no injuries were reported [1][2]. This limits the immediate legal exposure for Waymo but does not eliminate it. Had the vehicle struck a bystander, a pedestrian, or another car while entering or being swept through floodwaters, the liability analysis would become considerably more complex.

Product-liability law for autonomous vehicles remains unsettled. When a malfunction in software, sensors, or system design contributes to an accident, manufacturers and technology developers may face claims under strict liability for defective design, manufacturing defects, or failure to warn [15][16]. Liability can extend beyond the primary manufacturer to software developers, component suppliers, and fleet operators [16].

A distinct challenge for AV litigation is data access. Investigating an autonomous vehicle incident requires sensor logs, system status records, and decision-tree histories — data controlled by the manufacturer [15]. Courts and regulators are still working out who "owns" this data and under what circumstances plaintiffs can compel its disclosure.

Effective July 1, 2026, California Assembly Bill 1777 will authorize peace officers to issue "Notices of Autonomous Vehicle Noncompliance" directly to manufacturers for observed traffic violations [15]. This represents a shift toward treating AV operators more like human drivers in terms of real-time enforcement — though enforcement mechanisms for edge cases like flood driving remain undefined.

The Overreaction Argument

Critics of the recall's framing argue that the response is disproportionate. Human drivers routinely drive into flooded roads — it is the leading cause of flood-related deaths in the United States, killing an average of 88 Americans annually according to the National Weather Service. NHTSA does not recall human-operated vehicles after individual drivers make the same error.

The steelman case against treating this recall as a serious indictment of Waymo's technology runs as follows: the vehicle was empty, no one was hurt, the company identified the problem within days, paused operations voluntarily, filed a recall within ten days, and is deploying a fleet-wide fix that will prevent recurrence across all 3,791 vehicles simultaneously. A human driver who makes the same mistake in a flood gets no software patch. Every other human driver remains just as likely to repeat the error.

Proponents of this view argue that subjecting autonomous vehicles to a higher standard than human drivers creates a perverse incentive structure. If a single edge-case incident — one in which the AV's overall safety record still vastly outperforms human drivers — triggers headlines, recalls, and regulatory scrutiny, it raises the bar for deployment in ways that keep a statistically safer technology off the road longer, resulting in more crashes and fatalities overall [12].

Waymo's own framing reflects this tension. The company chose to file a formal NHTSA recall rather than quietly deploying a background software update, a decision that signals regulatory seriousness but also invites the very scrutiny that AV advocates argue is counterproductive [6].

Permits, Expansion, and the Local Government Response

The San Antonio incident arrived at a sensitive moment for Waymo's expansion plans. The company launched public service in San Antonio on February 24, 2026 — less than two months before the Salado Creek incident [17]. San Antonio was part of an aggressive 2026 expansion that also included Dallas, Houston, and Orlando, with Waymo laying the groundwork for service in more than 20 cities [17][18].

By March 2026, Waymo had expanded its San Antonio service area to include the San Antonio International Airport — the first fully autonomous ride service to offer airport transportation in Texas [19]. The service covered more than 60 square miles, from North Star Mall to downtown, the River Walk, and the Henry B. González Convention Center [17].

The flooding incidents forced a conspicuous pause just as the company was building public trust in a new market. Waymo suspended all San Antonio passenger operations following the April 20 incident and did not announce plans to resume until May 10 [8][9].

No Texas regulatory body has publicly revoked or suspended Waymo's operating permits. Texas has among the most permissive autonomous vehicle laws in the country, with no state-level permit requirement for AV testing or deployment [18]. This regulatory environment was a factor in Waymo's decision to expand into multiple Texas cities simultaneously [18].

The contrast with other jurisdictions is instructive. Waymo continued to expand Houston operations during the San Antonio pause, growing its Houston service area to nearly 50 square miles on May 14, 2026 — the day after the recall was publicly announced [20]. This suggests that neither NHTSA nor Texas state authorities viewed the flooding deficiency as grounds for a broader operational halt.

Timeline and Transparency

The sequence from incident to public disclosure raises questions about the speed of Waymo's safety review process:

• Early April 2026: First flooding incident near McCullough Avenue and Contour Drive in San Antonio [2]
• April 20, 2026: Second incident — vehicle swept into Salado Creek [1][2]
• April 20, 2026: Waymo deploys interim operational controls and weather restrictions [6]
• April 20–21, 2026: Waymo pauses San Antonio operations [9]
• April 30, 2026: Waymo files voluntary recall with NHTSA [2]
• May 10, 2026: Waymo announces plans to resume San Antonio service [8]
• May 12–13, 2026: Recall publicly reported [3][7]

The ten-day gap between the incident and the NHTSA filing is relatively fast by automotive recall standards. But the first flooding incident — the stranding near McCullough Avenue — occurred roughly two weeks earlier without triggering a recall. Whether Waymo's internal safety review should have escalated after the first event, rather than waiting for the more dramatic Salado Creek failure, is an open question.

NHTSA's Early Warning Reporting (EWR) system requires manufacturers to report certain safety-related data quarterly, but the reporting thresholds and timelines were designed for traditional automakers producing millions of vehicles, not fleet operators with fewer than 4,000 [13]. Whether this framework provides adequate early-warning capacity for AV-specific failure modes — like software that treats floodwater as navigable — remains untested.

What Comes Next

The Waymo flood recall is a relatively small action by the standards of automotive safety — 3,791 vehicles, no injuries, a software fix that requires no dealership visits. By the numbers, it is dwarfed by Tesla's 2.4-million-vehicle Autosteer recall and by the tens of millions of vehicles recalled annually for hardware defects like faulty airbags and brake failures.

But its significance lies in what it reveals about the gap between autonomous vehicle capability and the regulatory infrastructure governing it. Federal safety standards were written for cars with steering wheels and brake pedals operated by licensed humans. The SELF DRIVE Act of 2026 attempts to close that gap, but the bill has not yet passed, and the liability framework for software-driven decisions remains a patchwork of state tort law and ad hoc NHTSA enforcement actions [13][15].

Waymo's decision to treat the flooding deficiency as a formal recall — rather than a routine software update — sets a precedent that the company and its competitors will have to live with. Every future OTA patch that addresses a safety-relevant behavior will face the question: is this a recall, or just an update? The answer may depend less on engineering than on how much political and public pressure the industry can absorb while the rules are still being written.