NHTSA Tracked Every Robotaxi Crash for Four Years. On June 16, It Stops.

In June 2021, NHTSA did something unusual for a federal agency that regulates an industry worth trillions of dollars: it built a surveillance system that actually worked. Its Standing General Order required every manufacturer and operator of automated driving systems and Level 2 driver-assist vehicles to report every crash, to every relevant entity, within 24 hours of learning about it.^[1] That order produced the most comprehensive autonomous vehicle crash dataset any government has ever assembled. In eleven days, the replacement takes effect, and it guts the apparatus that built that dataset piece by piece.

The amended SGO, announced April 24, 2025 as part of DOT Secretary Sean Duffy's Automated Vehicle Framework,^[3] rewrites the reporting architecture in four fundamental ways. First, it eliminates the requirement that all reporting entities independently file on the same crash. Under the old rules, if a Waymo robotaxi collided with a delivery truck, both Waymo and the vehicle's operator needed to file separate reports, each offering their own narrative and data. Under the new rules, one entity files and the rest stay silent unless they possess "materially different information."^[4] The practical result is that NHTSA will receive one company's version of events for most crashes, and the agency will need to affirmatively request anything else.

Second, the two-tier reporting timeline collapses into a single, slower window. Severe crashes involving fatalities, hospital injuries, or vulnerable road users previously triggered a 24-hour reporting deadline and a mandatory ten-day follow-up with updated details. Under the amended SGO, everything merges into a five-day window with no follow-up requirement at all.^[4] Those ten-day follow-up reports mattered because initial filings are chaotic, incomplete, assembled from fragments of sensor data and insurance claims. That follow-up was where the picture sharpened. Eliminating it means NHTSA gets the blurry first draft and nothing more.

Third, minor property-damage crashes below $1,000 become effectively invisible unless the AV was the only vehicle involved or struck another vehicle or object.^[5] That threshold sounds reasonable until you consider that the minor-crash data is where patterns live. A robotaxi that clips mirrors in three parking lots across two weeks is not a single-crash problem. It is a systematic perception failure that the $1,000 cutoff would scatter into statistical noise before anyone could connect the dots.

Fourth, monthly status reports are now required only when "materially new or materially different information" exists for specific enumerated fields.^[5] Under the old regime, monthly reports created a standing cadence that kept manufacturers engaged with their own data. The new approach treats silence as the default, which is precisely what every regulated entity prefers.

The timing lands like a punchline written by someone who does not understand irony. Waymo published its 170-million-mile safety analysis in March 2026, reporting 92% fewer serious-injury crashes, 83% fewer airbag deployments, and 82% fewer injury crashes compared to human drivers in the same conditions.^[6] Those numbers are impressive and likely accurate. They are also derived from exactly the kind of comprehensive, multi-party, mandatory crash reporting that the amended SGO dismantles. The industry's strongest safety argument was built on a foundation the industry's lobbying arm just convinced the government to jackhammer.

Meanwhile, the street-level data tells a messier story. In Austin, the NTSB confirmed that a Waymo robotaxi illegally passed a stopped school bus while students were boarding, and that the decision was directed by a human remote-assistance agent.^[7] Another Waymo vehicle blocked an ambulance responding to a mass shooting on West Sixth Street. Since their Austin launches, Waymo has logged 60 reported incidents across 200 vehicles, while Tesla has recorded 15 across fewer than 40.^[7] In San Francisco's District 3, 311 complaints about autonomous vehicles are up 118% year-over-year, climbing from 11 to 24 in the same period.^[8]

Secretary Duffy framed the changes as removing unnecessary burden. "The long-term goal here is to move closer to a single national standard," he said. "We don't want 50 states with 50 standards."^[3] NHTSA Chief Counsel Peter Simshauser called them "the first steps toward making America a more welcoming environment for the next generation of automotive technology." Neither addressed the question that matters: if the data collected under the old rules demonstrated that autonomous vehicles are safer than human drivers, what happens to that claim once you stop collecting the data?

Limitations

California's DMV collision reporting program operates independently of the federal SGO and is unaffected by the amended rules. States with their own AV testing frameworks may implement compensating requirements, though most have not. Waymo's Swiss Re insurance analysis, covering 25.3 million miles with 88% fewer property-damage claims and 92% fewer bodily-injury claims, provides a verification path that does not depend on government reporting.^[6] The amended SGO is a real regulatory weakening, not a data blackout, but it shifts the burden from mandatory disclosure to voluntary transparency and government investigation.

The strongest case for the new rules

Requiring every entity to independently report the same crash was genuinely duplicative: NHTSA received multiple conflicting narratives of identical incidents, and sorting them consumed investigative bandwidth that could have gone to targeted probes. A streamlined, higher-quality dataset from a single designated reporter, combined with NHTSA's authority to request additional records on demand, could in theory produce better intelligence with less noise. The industry may also be right that the comprehensive reporting model cannot scale to millions of AV miles per week without becoming an exercise in bureaucratic paperwork rather than safety analysis. But "in theory" and "could" are doing significant structural work in that sentence, and the gap between them is where the crashes that nobody reports will happen.

What you should do

If you drive, walk, or cycle in a city with active robotaxi deployments, the federal government is about to know less about what happens when those vehicles crash. File complaints with NHTSA's Vehicle Safety Hotline and your city's 311 system. California residents can check collision reports at the CA DMV, which still requires ten-day reporting. And if a robotaxi hits your car for under $1,000 in damage after June 16, understand that under federal rules, nobody is required to tell NHTSA it happened.