The National Highway Traffic Safety Administration opened Preliminary Evaluation PE26007 on September 21, 2026, into comma.ai’s openpilot driving-automation hardware after identifying five crashes in which the system failed to detect stopped or slow-moving vehicles directly in its own lane. Three of those crashes killed a total of three people; the other two injured 11 more, some seriously. The device at the center of the inquiry sells for under $1,000 and is designed to be installed by the owner, not a dealer or a factory.
An estimated 30,000 of the units are already in service, wired into vehicles across more than 325 different car models, according to NHTSA’s own investigation filing.
How five crashes exposed the same blind spot
Investigators did not describe five unrelated incidents; they described a repeated pattern. In each case, a comma-equipped vehicle struck a vehicle that was stopped or moving slowly directly ahead, in the same lane. NHTSA’s Office of Defects Investigation resume puts the finding plainly: the agency suspects “the comma system may not have adequately detected or responded to the in-lane vehicles” it was supposed to be watching for. A generic autonomy complaint might describe a system that simply steers poorly; this filing describes something narrower, a specific and repeatable blind spot for traffic sitting directly in the vehicle’s own path.
Two of the five crashes were fatal, together killing three people; the other three crashes injured 11 more, some of them seriously, per the same filing.
A device built to work on almost any car
What distinguishes openpilot from the driver-assistance systems built into new vehicles is that no one has to buy a new vehicle to get it. comma.ai’s own site markets the device as hands-free driving for a car an owner already has, with a self-install the company says takes about 15 minutes and no subscription required afterward, running on openpilot’s open-source code. The company describes the intended use as “eyes-on” rather than fully autonomous, meaning drivers are still expected to watch the road even while the system steers and brakes on their behalf.
An aftermarket price point under $1,000, paired with compatibility claimed across more than 325 models, is what let the installed base reach an estimated 30,000 units without a single automaker’s cooperation or a single factory certification process behind it. Electrek’s reporting on the filing notes the devices have logged more than 400 million miles of real-world driving between them, a scale that makes the five-crash count NHTSA is investigating a comparatively small fraction of total use, though the fatalities and injuries within that fraction are what triggered the inquiry. Unlike a factory driver-assistance suite, which a single automaker can update fleet-wide overnight, an aftermarket device’s fixes depend on individual owners downloading and installing new software themselves.
comma.ai’s response, on the record
comma.ai has not disputed that the crashes happened. According to Self Drive News, which first detailed the filing, the company has told NHTSA it is providing the requested crash information “without agreeing with NHTSA’s interpretations or conceding that comma is obligated to respond,” a carefully hedged cooperation that stops short of admitting fault while not withholding data either.
Such a posture is not unusual for a Preliminary Evaluation, but it leaves the central question unresolved for now: whether the pattern NHTSA identified reflects a software limitation openpilot shares across its installed base, or five instances of drivers over-relying on a system the company’s own marketing describes as requiring eyes on the road at all times. Nothing in the filing NHTSA published assigns blame between those two explanations; it only documents that the same failure showed up five separate times.
What happens next inside a Preliminary Evaluation
A PE is an early stage, not a conclusion. Under NHTSA’s own description of the process, an investigator reviews existing evidence to determine whether it supports “a finding of an unreasonable risk to motor vehicle safety,” and from there the case either closes for lack of sufficient evidence, escalates into a more formal Engineering Analysis with testing and further information requests, or ends with the manufacturer filing what the agency calls an influenced recall.
NHTSA’s filing does not name the three hardware variants involved by crash, only that the installed base spans comma three, comma 3X and comma four units, each running the same underlying openpilot software rather than separate code per device generation. That detail matters for how far any eventual fix would have to reach: a software-level cause would implicate the full 30,000-unit population regardless of which generation of hardware a driver happens to own, while a hardware-specific fault would narrow the fix to whichever variant sold most heavily during the period the five crashes occurred.
No timeline exists for which of the three paths PE26007 will take. What is fixed, for now, is the count NHTSA opened with: five crashes, three deaths, 11 injuries, and an estimated 30,000 of the devices already on the road while the agency decides.
This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.
More from Morning Overview
- A handful of car engines are so tough mechanics say they almost never wear out
- The NSA is again telling phone owners to switch off one location setting
- Supplements now rank as the fifth-leading cause of death from liver disease.
- Herbal supplements are landing Americans in the hospital with liver damage, doctors warn
