NHTSA Investigates openpilot Assist System

The openpilot driver assistance software is now the target of a formal federal safety investigation. The National Highway Traffic Safety Administration (NHTSA) opened the probe after five crashes where vehicles using the aftermarket system crashed into stopped or slow-moving vehicles in their own lane. These incidents were severe. Two of those crashes killed three people, and four of the collisions left 11 other people injured, some seriously.

The investigation is a massive hurdle for comma.ai, the startup founded by George Hotz. The federal probe covers an estimated 30,000 devices, specifically targeting the last three generations of the company's hardware: the comma three, comma 3X, and comma four. This is the biggest regulatory crisis for the startup since 2016, when federal regulators pressured the company to cancel its very first product.

The Core of the Federal Case

The Office of Defects Investigation opened a Preliminary Evaluation on September 21. It's called PE26007. The investigation targets the physical hardware devices alongside the open-source software, openpilot, that powers them, and while that may sound like a narrow technical focus, the scope here reaches every driver who's ever trusted a system they don't fully understand. Investigators gathered data on these crashes through consumer complaints, direct outreach to police departments, and crash reports filed under its Standing General Order. And they're not done.

So what does this actually mean for you? Federal investigators have confirmed the system was active during multiple crashes. Their early review of the data suggests that the software may not have adequately detected or responded to the in-lane vehicles. It's the core issue. And this gets to the very heart of the safety debate surrounding aftermarket driver assistance technology, a debate that's still unfolding because we've got questions that nobody, not the regulators and not the companies selling these systems, can fully answer yet.

The Disconnect in Marketing

Federal regulators are looking closely at how the technology is marketed versus how it actually functions. They're asking hard questions. The investigation highlights a sharp contrast between promotional claims and the legal fine print, a gap that matters because what a company says in its ads and what it buries in the terms of service can be two very different things. And the startup's marketing materials have promised:

• Hands-free driving for the car you already have
• A quick 15-minute self-installation process
• Software that can drive for hours without intervention

But there is a catch. Despite these bold marketing claims of hands-free driving, the company's fine print tells a completely different story. The user agreements and product documentation state that the driver must remain fully alert and ready to take over control of the vehicle at all times.

What the Crash Reports Reveal

The public crash data shows a pattern. It's repetitive. It's deeply concerning. In February 2026, in Ascension Parish, Louisiana, a 2022 Toyota RAV4 equipped with a comma device struck a stopped first responder vehicle on a highway, and then in September 2025, a 2020 Honda Accord rear-ended a stopped car on a Missouri highway at 82 mph, sending an individual to the hospital with serious injuries. We've seen this before. And in July 2026, a 2021 Toyota Corolla hit a stopped passenger car at 56 mph on a highway near Mona, Utah.

Cut the noise. Every single one of these crashes involved highway speeds and a stationary vehicle sitting directly in the lane ahead. Stationary vehicles are a notorious weak spot for camera-based driver assist systems, a technical challenge that has plagued other major automakers for years, and it's a problem they're still dealing with. They can't crack it. And we've seen the same pattern repeat, crash after crash, because a camera alone just don't catch what's sitting still in front of it at speed.

The Battle Over Responsibility

The most telling details of this investigation emerge from how the company reports these incidents. It's all in the filings. In its official filings, the startup includes a standard disclaimer with every single crash report, and that disclaimer shows up each time, without exception, attached to each report the company submits to the government. The company states it provides the information without agreeing with federal interpretations or conceding that it is legally obligated to respond to the government's General Order. And they don't concede a thing.

The company often says "Unknown." Asked whether crashed vehicles ran within proper design parameters, it frequently gives that same answer. And they're clear about why: they don't manufacture the actual vehicles or the stock driver assistance systems built into them, so the question, in their view, isn't theirs to answer. In the Missouri Accord crash, the company added a note in capital letters stating that adaptive cruise control was performed by the Honda system.

"appears to have been running 3rd party non-comma software"

The company put it in its report. That statement concerns the fatal Louisiana crash. It highlights the complex legal and technical battle ahead, and it's a battle that touches on questions of liability, software modification, and what any automaker or developer can reasonably be expected to control once its code leaves its own hands.

The vehicle was reportedly running a fork. It's a modified, third-party version of the openpilot software. And they're not the only ones asking what that means.

The Problem With Open Source Forks

It's open source. That makes this investigation incredibly complicated. Many users don't run the stock software, and instead they use popular community-developed forks like sunnypilot and FrogPilot, which are built on the exact same driving model and code base. So the code is shared.

Regulators are already anticipating this defense. The safety agency stated that if a modified version of the software is involved in a crash, it will evaluate the system and focus on the shared components between the stock software and the third-party fork. If the core code is the problem, blaming third-party developers might not protect the company.

From Dev Kits to 30,000 Cars

This isn't the startup's first regulatory battle. In 2016, founder George Hotz made a choice that stunned the industry: rather than comply with a special order from federal regulators, he killed the original comma one product entirely. He just walked away. So to bypass those restrictions, the company open-sourced the openpilot software and began selling its hardware devices as "developer kits," a move that let it keep operating while regulators watched.

That pivot worked. It's a hit. The company now claims that more than 30,000 cars use its devices on the road today, and those devices have logged over 400 million miles, with the company claiming to be second only to Tesla in total hands-free miles driven. So the comma four hardware launched last November at a price of $999, and the company claims it's compatible with over 350 car models. We've got the numbers. They don't lie.

What Happens Next

Real talk: this investigation places the company in a very difficult position. You cannot market a product as a hands-free system that drives for hours, sell it for a quick DIY install, and then argue to federal regulators that you have no responsibility when a crash occurs. Those two positions are completely incompatible.

This preliminary evaluation could take months to resolve. It might end quietly. No action at all. Or it could escalate to a formal engineering analysis and a mandatory safety recall. With 30,000 devices active on public roads and three deaths already documented, executing a recall on open-source software that anyone can modify on GitHub presents an unprecedented regulatory challenge, one that regulators don't have a clear playbook for. But they're going to have to figure something out.

Frequently Asked Questions

What triggered the NHTSA investigation into the openpilot system?

The National Highway Traffic Safety Administration opened the probe after five crashes where vehicles using the aftermarket system crashed into stopped or slow-moving vehicles in their own lane. Two of those crashes killed three people, and four of the collisions left 11 other people injured, some seriously.

Which specific hardware devices are covered by the federal investigation?

The federal probe covers an estimated 30,000 devices, specifically targeting the last three generations of the company's hardware: the comma three, comma 3X, and comma four. The investigation targets the physical hardware devices alongside the open-source software, openpilot, that powers them.

What contrast does the investigation highlight between the company's marketing and its fine print?

The startup's marketing materials have promised hands-free driving for the car you already have, a quick 15-minute self-installation process, and software that can drive for hours without intervention. However, the user agreements and product documentation state that the driver must remain fully alert and ready to take over control of the vehicle at all times.

How does the company's open-source model complicate the investigation?

Many users don't run the stock software, and instead they use popular community-developed forks like sunnypilot and FrogPilot, which are built on the exact same driving model and code base. The safety agency stated that if a modified version of the software is involved in a crash, it will evaluate the system and focus on the shared components between the stock software and the third-party fork.

What possible outcomes could result from this preliminary evaluation?

This preliminary evaluation could take months to resolve, and it might end quietly with no action at all. Or it could escalate to a formal engineering analysis and a mandatory safety recall.