Regulators Are Investigating Openpilot After Five Crashes Killed Three

The National Highway Traffic Safety Administration has opened a formal investigation into comma.ai, the company that sells a plug-in box promising hands-free driving for cars that never shipped with it. The agency's Office of Defects Investigation opened Preliminary Evaluation PE26007 on September 21, according to Electrek, after identifying five crashes in which vehicles running comma hardware drove into stopped or slow-moving traffic in their own lane. Three people died across two of those crashes, and 11 others were hurt, some seriously.

The probe covers roughly 30,000 devices, per Electrek's reading of the filing: the comma three, comma 3X and the current comma four, which retails for $999. It also covers openpilot, the open-source software that runs on them. NHTSA's preliminary finding is blunt. Its early review of the data suggests the system may not have adequately detected or responded to the vehicles ahead, according to the agency language quoted by CarBuzz.

The crash pattern NHTSA is chasing

The five crashes share a signature that anyone who followed the Tesla Autopilot investigations will recognize: a car traveling at highway speed, a stationary or slow vehicle in the lane, and no meaningful braking. Electrek lists three of the incidents. In February 2026, a 2022 Toyota RAV4 in Louisiana struck a stopped first-responder vehicle; Runtime Wire reports that two occupants died. In September 2025, a 2020 Honda Accord in Missouri rear-ended a stopped car at 82 mph after the driver set cruise control near 80 and fell asleep. In July 2026, a 2021 Toyota Corolla hit a stopped passenger car at 56 mph near Mona, Utah.

Electrek also notes the parallel with the agency's 2021 Autopilot probe, which also centered on cars driving into stopped vehicles. That is the template regulators have for this failure mode, and it is why a relatively small company now finds itself in the same docket category as the largest EV maker in the country.

The open-source problem

What makes this case unusual is the software supply chain. Comma has a history with NHTSA: in 2016, the agency pressed founder George Hotz over his original product, the comma one, and he cancelled it. According to Carnegie Mellon's Safety21 center, the company responded by moving its driving software and hardware designs onto GitHub, where anyone can download, modify and build them.

That decision is now central to the investigation. Comma told NHTSA the Louisiana RAV4 appears to have been running third-party software rather than its own release, per Electrek. Runtime Wire reports that at least two of the five crashes involved community forks, including FrogPilot, a popular modified build with its own tuning and feature set. NHTSA is examining both the official release and the forks.

This is a question regulators have not had to answer at this scale. Traditional defect law assumes a manufacturer controls what code ships in a car. Openpilot inverts that. Comma sells hardware and publishes a reference build, but any owner can flash a fork that changes following distance, acceleration limits or how aggressively the system trusts its own perception. If a fork causes a crash, is the defect in comma's device, in the upstream code the fork was built on, or in the fork itself? A preliminary evaluation is the first step toward an engineering analysis and possibly a recall, so how NHTSA draws that line will matter well beyond one company.

Marketing versus the fine print

The agency's filing also leans on how comma sells the product. NHTSA quotes the company's pitch of hands-free driving for the car you already have, a 15-minute self-install, and software that can drive for hours without intervention, according to Electrek. The company's own terms say the opposite in practice: openpilot requires the driver to be alert and ready to take over at all times, and the system includes a driver-monitoring camera to flag distraction, as comma emphasized to CarBuzz.

That tension between the headline promise and the legal disclaimer is exactly what regulators scrutinized in Tesla's case. It matters more here because comma's customers install the device themselves, on cars from more than 350 supported models per Electrek and CarBuzz, and can swap in a fork with no dealer or automaker in the loop.

Comma's posture so far has been combative in the paperwork. Electrek reports that the company's filings include language stating it provides information without agreeing with NHTSA's interpretations, and that it has argued braking responsibility sits with the vehicle manufacturer, since openpilot sends commands through each car's own control systems.

Why it matters

Comma claims its fleet has logged more than 400 million miles and says it trails only Tesla in hands-free miles, per Electrek. Flux's math: five crashes over that mileage is not, by itself, evidence that openpilot is less safe than a human driver. But defect investigations are not actuarial exercises. They ask whether a specific, repeatable failure exists, and stopped-vehicle detection is the same weakness that put Tesla's Autopilot under federal scrutiny.

The broader stakes are about the aftermarket. Level 2 retrofit kits sit in a gray zone: they are not built by automakers, they ride on factory safety systems they did not design, and their software can be rewritten by users overnight. If NHTSA escalates this to a recall, it will have to decide whether a recall can even apply to code that anyone can fork. If it closes the probe, it effectively tells the retrofit market that open source is a liability shield. Either way, the outcome will set the rules for everyone trying to bolt autonomy onto cars that were never designed for it, from hobbyist kits to the startups pitching fleet retrofits for robotaxi service.

For now, the investigation is early. A preliminary evaluation is not a finding of defect, and no recall has been ordered. But the agency has named the failure mode, counted the dead, and put the open-source model itself on the table.