Three people are dead and eleven more were injured in crashes involving vehicles equipped with aftermarket driver-assistance hardware. The National Highway Traffic Safety Administration is now examining whether a $999 retrofit device that provides steering and speed-control assistance contributed to those outcomes.
NHTSA opened a preliminary evaluation on September 21, 2026, targeting Comma AI’s hardware after five reported crashes involving vehicles equipped with the company’s devices. Preliminary information indicated the systems may not have adequately detected or responded to stopped or slowly moving vehicles in the same travel lane.
This investigation does not only affect drivers who chose to install the hardware. Every driver sharing a lane with one of these vehicles has a stake in what NHTSA finds.
A $999 Device, 300-Plus Compatible Vehicles, and an Unanswered Question
The gap between how Comma AI markets its hardware and what its own safety language requires is at the center of what regulators must now assess.
Comma AI sells a device called the Comma Four for $999. It is compatible with more than 300 vehicle models and runs openpilot, an open-source software system capable of performing adaptive cruise control and automated lane centering, including steering and adjusting speed in response to traffic conditions.
Comma’s own safety language requires the driver to remain fully alert and ready to retake control immediately. Openpilot is a driver-assistance system; the human at the wheel remains responsible for the vehicle at all times.
That distinction matters. Comma’s marketing emphasizes hands-free operation, while its safety requirements demand continuous driver attention. When those two messages compete, the one that sounds like freedom tends to win.
The deeper regulatory question is not simply whether the hardware works as described. It is who bears responsibility when a consumer-installed system, running software the manufacturer did not write or validate, influences a vehicle’s behavior on a public road and something goes wrong.
The Fork Problem Regulators Are Now Trying to Untangle
Because Comma’s hardware openly supports community-modified software, investigators cannot easily determine which version of openpilot was running at the moment of any given crash.
Openpilot is open-source software, which means outside developers can copy the underlying code, modify it, and maintain their own versions. Those versions are called forks. Comma Four explicitly supports community forks, meaning users are not limited to the software Comma ships.
Sunnypilot is one such fork. A publicly shared demonstration, reported by Futurism, showed a driverless Tesla Model Y operating with Comma Four hardware running Sunnypilot rather than Tesla’s own Full Self-Driving system. The person sharing the demonstration reportedly explained that their Tesla FSD subscription had lapsed and that the vehicle was using an open-source alternative instead.
NHTSA’s evaluation is examining both the Comma hardware and the possible role of modified software in the reported crashes. A fork is not inherently unsafe, but it creates a verification problem that factory-developed systems generally do not present.
Determining which software build was running, and in what configuration, at the moment of a collision is a challenge the agency’s investigation will need to address. That uncertainty affects not only crash investigators but also insurers and anyone trying to understand what actually occurred.
One commentator, Cairo Smith, captured a widely shared public reaction in remarks reported by Futurism: “I do not think we should be letting people run any software they want on their self-driving car.” That is an opinion, not a safety finding, but it reflects the accountability gap regulators are now being asked to address.
The open-source software culture that built Linux and made browser extensions routine operates on an assumption that modification carries manageable risk. Applying that assumption to systems controlling a vehicle’s steering and speed on roads shared with people who never consented to participate is a different proposition.
What NHTSA Is Doing and What It Has Not Concluded
The agency’s preliminary evaluation establishes a crash pattern serious enough to warrant scrutiny, not a determination that the hardware caused the reported deaths.
A preliminary evaluation is not a defect determination, and it does not establish that Comma hardware or openpilot caused the reported crashes. It does establish that a recurring pattern of incidents was serious enough to require federal examination.
According to NHTSA’s evaluation document, the agency plans to verify system status in each of the five incidents and evaluate the systems’ performance and safety safeguards. It will also examine whether modified software played a role in any of the collisions.
Two of the five crashes resulted in three fatalities. Four crashes involved 11 additional people who suffered nonfatal injuries, including serious ones.
The crash pattern NHTSA identified is specific: vehicles reportedly failed to detect or respond to stopped or slow-moving vehicles already in the same lane. Detecting a stopped or decelerating vehicle ahead is among the core tasks a driver-assistance system must perform reliably.
What follows from this investigation may extend well beyond Comma AI. Regulators could require more detailed software-version disclosures in crash reports and tighter restrictions on how driver-assistance systems are marketed. Clearer safeguards governing which software modifications are permissible on vehicle-control hardware may follow as well. The open-source norms that work for apps and operating systems were not designed with public-road safety consequences in mind. NHTSA is now tasked with deciding whether the rules need to catch up.
