comma sells a $999 box that sticks to your windshield, plugs into your car’s wiring and promises hands-free highway driving. Federal regulators now want to know why some cars wearing one ran into traffic that was stopped or crawling in their own lane.
NHTSA’s Office of Defects Investigation opened Preliminary Evaluation PE26007 on September 21, 2026, covering the comma three, comma 3X, comma four and the openpilot software, with an estimated 30,000 units in scope. The opening resume lays out the pattern: five crashes in which comma-equipped vehicles reportedly hit stopped or slow-moving vehicles in their own lane. Two of those crashes killed three people, and four of the five injured eleven others, some seriously.
NHTSA’s own summary puts it this way: early data suggests the system “may not have adequately detected or responded to the in-lane vehicles.”
What the paperwork actually says
This isn’t a probe built on one angry owner. ODI lists three triggers: crash reports filed under Standing General Order 2021-01, a vehicle owner complaint, and outreach to police departments. The five crash reports are public and can be pulled from NHTSA’s SGO data page by ID number.
Related Articles
- Pirelli Is Dropping $1.2 Billion in Georgia on Robot-Built Tires That Talk to Your Car
- Car Loan Rates Look Cheap Right Now. The Bond Market Says That Won’t Last.
Investigators say they’ve confirmed that the comma system was powered on or engaged in multiple crashes. Note the word “multiple,” not “all.” Establishing exactly what was running, and whether it was in control, is the first job of this evaluation. ODI says it will pull data to verify system status in each incident and assess both performance and the safeguards meant to keep equipped vehicles operating safely.
Why stopped cars are the nightmare scenario
A car cruising at 70 mph closing on a stationary one is the hardest case for these systems, and it is also the most lethal.
The engineering reason is simple. Traditional radar-based adaptive cruise systems see a world full of stationary objects: signs, guardrails, overpasses, parked cars on the shoulder. To avoid slamming on the brakes for every bridge, they tend to discount things that aren’t moving. That’s why many owner’s manuals for factory systems carry fine print warning that the system may not respond to stopped vehicles. A car that suddenly appears when the vehicle ahead changes lanes is the classic trap.
comma’s approach layers its own camera-based perception on top of the car’s factory hardware. NHTSA describes the devices as connecting to existing vehicle systems through wire harnesses and using the device’s cameras together with stock vehicle sensors to provide sustained automated control of speed and steering. That hybrid design matters for the investigation. With a system that blends aftermarket vision and factory radar and actuators, figuring out which component failed to react is a hard forensic problem. Was it comma’s perception, the factory cruise logic, or the handoff between the two? That question could define where responsibility lands.
Marketing vs. the fine print
The NHTSA resume doesn’t just describe the hardware. It describes the sales pitch. The agency points out that the devices are sold directly to consumers with do-it-yourself installation instructions, and that the company advertises a 15-minute setup and software that can drive for hours without intervention.
Visit the comma homepage today and the pitch is “hands free driving for the car you already have,” alongside claims of more than 30,000 cars on the road, over 404 million miles driven, and second place behind Tesla in hands-free miles. The comma four sells for $999 with a 30-day risk-free trial.
Now scroll to the bottom of the openpilot repository on GitHub. In capital letters, comma declares the software to be alpha-quality research software, adding “THIS IS NOT A PRODUCT.” The same page asks users to indemnify Comma.ai against claims arising from their use of the software.
That’s a hard line to hold: a $999 consumer device with a 30-day trial, running code whose own license says, in all caps, that it isn’t a product. To be fair, comma doesn’t hide the driver’s role. NHTSA itself notes that comma’s website says openpilot requires an alert driver ready to take over at all times. The company’s safety document goes further, stating that “driver alertness is necessary, but not sufficient” for safe use.
The open-source wrinkle
Unlike NHTSA’s probes into automaker systems, this one involves code anyone can modify. The main repository shows roughly 11,400 forks, and some community versions add features or loosen restrictions.
NHTSA is clearly aware of this. The resume says that when a modified version of openpilot is involved in a crash, ODI will evaluate it and consider the components that the fork shares with comma’s own software. That’s significant: comma can’t simply argue that a crash happened on someone else’s code if the shared building blocks are implicated.
comma has tried to police this. Its safety rules tell fork developers not to disable or weaken driver monitoring or the excessive-actuation checks, and forks that modify the core safety code must keep the full safety test suite passing. Noncompliant forks and their users can be banned from comma’s servers, and the company says it strongly discourages running forks with missing or incomplete safety code. A server ban stings a hobbyist. Whether NHTSA counts it as a safety control is a different question.
comma has walked away from NHTSA before
comma and NHTSA have met before. In October 2016, the company’s founder, George Hotz, responded to an agency letter about the original comma one by posting on comma’s X account: “Would much rather spend my life building amazing tech than dealing with regulators and lawyers. It isn’t worth it.” The product was cancelled.
Ten years and tens of thousands of devices later, walking away is not a realistic option. A preliminary evaluation is the first formal step in NHTSA’s defect process. It can close without action, or it can escalate to a deeper engineering analysis and, potentially, a recall push. Aftermarket equipment makers are not exempt from federal safety defect rules just because they don’t build the car.
What owners should know right now
Nothing in the file orders anyone to stop using their device, and no defect has been found. Still, owners should take a few practical steps.
Treat stopped traffic as a manual-driving event. If you see brake lights stacking up ahead, especially where a lead car might swerve out of the way, cover the brake and be ready to take over. That’s the precise scenario under investigation.
Be honest about your software. If you’re running a community fork, understand what it changes. Forks that modify safety logic put you outside comma’s own guardrails, and after a crash, the software version will be one of the first things investigators and insurers want to know.
Related Articles
- A Flock Camera Helped Put Her in Jail for 13 Days. Photos of Her Undamaged Durango Got Her Out.
- A Tesla Driver’s iPhone Called 911 on Its Own After He Crashed. Deputies Say He Didn’t Stick Around.
Remember the car is recording. By default, openpilot uploads driving data to comma’s servers, logging road-facing camera footage, CAN bus data, GPS and crash information. In a collision, that data could help your case or sink it, and it may be subpoenaed.
Call your insurer. Policies vary widely on aftermarket driver-assist hardware, and an undisclosed modification that affects vehicle control is not something you want an adjuster to discover after a claim. A five-minute phone call now beats a coverage dispute later.
Watch the docket. NHTSA updates investigation files as they progress, and any change in PE26007’s status will show up there before it shows up anywhere else.
If you run openpilot or a fork: has it ever hesitated on stopped traffic, and what did you do? Would a recall change whether you keep using it? Tell us in the comments.