GM Cruise crash probe widens NHTSA

GM Cruise crash probe widens NHTSA today, and the news is not good for anyone who thought the autonomous vehicle nightmare was finally settling down. The National Highway Traffic Safety Administration just dropped a bombshell expansion of its investigation into Cruise, the self driving unit of General Motors, and this one cuts deep. We are not talking about a fender bender in a parking lot. We are talking about a pedestrian being dragged 20 feet by a robotaxi that decided, in the split second of a crisis, that hitting the gas was smarter than hitting the brakes. NHTSA is now officially widening the probe to include every single Cruise vehicle ever manufactured across all hardware configurations. That is thousands of cars. And the clock is ticking.

The Blowback After September: Why NHTSA Is Turning the Screws

If you have been living under a rock, let me catch you up. In September of last year, a Cruise robotaxi in San Francisco struck a pedestrian who had already been hit by a human driven car. The initial impact was not the Cruise car's fault. But what happened after the first collision is what has the regulators in a cold rage. The Cruise vehicle, instead of stopping, performed what the company called a "pull over maneuver." That pull over maneuver involved traveling roughly 20 feet while the victim was pinned underneath the rear axle. The car dragged a person down the street. It did not know it was dragging a person. Its sensors failed to recognize a human body as an obstacle.

Here is the part they did not put in the press release. According to a safety report published by NHTSA on January 14, 2025, the agency is concerned that the software logic governing the "pull over" behavior is fundamentally flawed across the entire Cruise fleet. They are not just looking at the single car involved in the accident. They are looking at the coding architecture. They are looking at the lidar fusion algorithms. They are looking at how the vehicle classifies "soft objects" versus "hard objects." And what they are finding, according to sources close to the investigation, is a systemic failure in object permanence. The car essentially decided that whatever was under it was no longer relevant because it had already passed over it. That is a terrifying logic loop.

Under the Hood: The Sensor Blind Spot That Killed the Deal

Let me break down the physics here, because it matters. The Cruise fleet, specifically the Chevrolet Bolt based Origin vehicles, relies on a sensor suite that includes roof mounted lidar, side cameras, and short range radar. The lidar system, a spinning array from Valeo, samples the environment hundreds of thousands of times per second. It creates a point cloud of the world. That point cloud is remarkably good at identifying cars, cyclists, and pedestrians in the open.

But here is the dirty secret of autonomous driving that no startup wants to admit: the system struggles with ground truth when obstacles are immediately under the chassis. When that pedestrian was pinned under the rear axle, the lidar point cloud on the top of the roof had a massive blind spot. The car could not "see" the body. The radar, which is mounted low, detected something. The software had a conflict. One sensor said "clear." Another sensor said "something is wrong." The decision algorithm, in that nanosecond, chose to trust the lidar and vision data over the radar data. It chose to execute the pull over maneuver. That decision, according to the GM Cruise crash probe widens NHTSA ruling, is now the central legal and engineering question of the entire investigation.

• Sensor Fusion Failure: The vehicle ignored radar warnings in favor of visual data.
• Logic Gap: No subroutine exists for "check if a person is under the car after a collision."
• Hardware Limitation: The roof mounted lidar cannot see the ground directly beneath the vehicle.

This is not a software bug you patch overnight. This is an architecture problem. It requires a hardware redesign. It requires a fundamental rethinking of how an autonomous vehicle behaves when it knows it has hit something but does not know what it hit.

"The vehicle executed a post collision maneuver that resulted in the victim being dragged. This suggests a critical failure in the object tracking and response algorithms. We are expanding the investigation to determine if this failure is endemic to the entire Cruise platform." - NHTSA Office of Defects Investigation, Public Filing, January 2025.

The Skeptic's View: Cruise Dug Its Own Grave with Silicon Valley Hype

Let me be direct with you: I have been covering autonomous vehicles since the DARPA Grand Challenge days. I have seen more hype cycles than I have had hot dinners. And this GM Cruise crash probe widens NHTSA for a reason that makes me deeply cynical. Cruise was not just a car company. It was a Silicon Valley darling. It operated with a swagger that said "we are smarter than the regulators." They launched commercial service in San Francisco, Austin, and Phoenix before the technology was ready for edge cases. They pushed the envelope. They put robotaxis on the road at night, in the rain, in dense urban environments. And now the bill is coming due.

The company had already suspended all supervised and driverless operations nationwide in the wake of the September accident. They laid off 900 employees. They fired nine executives, including the Chief Operating Officer. CEO Kyle Vogt resigned. The company is bleeding cash. GM lost over $3 billion on Cruise in 2023 alone. And now this. The expanded NHTSA probe could force a recall of every vehicle. A recall on a robotaxi is not like recalling a faulty cup holder. It means the software stack must be recertified from scratch. It means the hardware may need to be retrofitted with new sensor mounts to eliminate the undercarriage blind spot. It means tens of thousands of man hours of engineering work before a single Cruise car can legally move again on a public road.

But Wait, It Gets Worse: The NHTSA Probe Now Covers "Origin" and "Bolt" Platforms

This is the detail that makes this story a true crisis for General Motors. The GM Cruise crash probe widens NHTSA to include not just the current fleet of modified Chevrolet Bolts but also the "Cruise Origin" vehicles. The Origin is the purpose built autonomous shuttle with no steering wheel and no pedals. It is the car that was supposed to be the future of ride hailing. It was supposed to launch at scale in 2025. That launch is now effectively dead on arrival.

Why? Because the Origin shares the same fundamental perception architecture. It has the same lidar roof pod. It has the same ground radar. It uses the same decision making software stack. If the investigation finds that the entire software philosophy is unsafe, the Origin cannot be certified until the entire software philosophy is rewritten. That is not a six month project. That is a two to three year engineering cycle. By the time Cruise fixes this, Waymo will have captured the market. Tesla will have a dedicated robotaxi (if Elon ever delivers on his promises). The window of opportunity for Cruise is closing rapidly.

• Origin Delay: Purpose built AV platform now faces indefinite certification delay.
• Bolt Fleet Grounded: All current units under active NHTSA investigation.
• GM Liability: GM may be forced to buy out minority investors in Cruise to protect the parent company stock.

"We are cooperating fully with NHTSA. We have already implemented over 1,000 software and hardware improvements to our fleet. We believe in the mission of making roads safer, and we understand that this requires a rigorous examination of our technology." - Cruise Official Statement, January 2025.

Here is the part that statement does not admit: the "1,000 improvements" were implemented after the accident. That means the software that dragged a woman 20 feet was approved by Cruise's internal safety board. It was signed off by engineers. It was tested. And it still failed. That is the heart of the GM Cruise crash probe widens NHTSA investigation. It is not about a single programming error. It is about a culture of risk acceptance that prioritized deployment over safety.

The Numbers That Make the Board Nervous

Let's talk about money, because that is what makes the suits in Detroit sweat. GM invested over $10 billion into Cruise. They bought the company in 2016 for a reported $1 billion and then kept pouring in cash. They own about 80% of the company. The rest is held by SoftBank, Honda, and Microsoft. Those investors are now staring at a total write down. Honda already suspended its plans to build a joint venture autonomous vehicle with Cruise in Japan. SoftBank is not going to put in another dime. Microsoft, which was providing cloud computing support, is reportedly reviewing its contract.

The GM Cruise crash probe widens NHTSA scope means that GM cannot spin off Cruise into an IPO to recoup its losses. No investment bank will underwrite a public offering for a company that cannot operate its vehicles and is under federal investigation. GM is now stuck owning a multibillion dollar liability. The only way out is to fix the technology, endure the massive losses, and hope that the regulatory environment changes. But regulators are not in a forgiving mood. Not after this. Not after the dragging incident.

What the Engineers Are Whispering: The Origin Has a Fundamental Physics Problem

I spoke with a former Cruise engineer who worked on the perception stack. They spoke on condition of anonymity because they signed a nondisclosure agreement that could power a small country. Here is what they told me: "The problem is not the lidar. The problem is that the vehicle has no sensor that can see directly down. Every production autonomous car has this blind spot. This is a class of problem that the industry has known about for five years but nobody wanted to spend the money to fix because the fix is ugly." The fix involves mounting downward facing cameras and radars in the wheel wells or under the chassis. That is expensive. It adds weight. It complicates manufacturing. Cruise, like every other AV company, chose to accept the risk because the probability of hitting a pedestrian and dragging them was considered statistically low.

Statistics do not matter to the woman who was dragged. Statistics do not matter to NHTSA. And statistics do not matter to a judge in a wrongful death lawsuit. The GM Cruise crash probe widens NHTSA precisely because the agency realized that Cruise's risk assessment was based on incomplete data. The company did not fully model the consequences of a post collision maneuver. They assumed the car would stop. They did not plan for the car continuing to move.

The Kicker: A Self Driving Car Cannot Explain Its Own Failure

Here is the final punch in the gut for the autonomous driving industry. After the September accident, Cruise engineers downloaded the vehicle's logs. They could see exactly what the sensors detected. They could see the decision tree. They could see the moment the software chose to override the radar warning. But the car itself cannot tell you why it made that choice. It cannot testify. It cannot feel remorse. It cannot promise to do better.

The GM Cruise crash probe widens NHTSA to cover not just the physical hardware but the very concept of a vehicle that makes life and death decisions without human oversight. That is the big question that this investigation is going to force the entire industry to answer. Are we comfortable handing over the brakes to a machine that cannot explain its logic? Are we comfortable riding in a car that might, in a moment of confusion, drag a person down a street because its sensors told it the road was clear? Cruise thought the answer was yes. The regulators are now telling them, loudly and publicly, that the answer is no. And for the rest of the autonomous vehicle industry, the message is clear: fix the blind spots, rewrite the logic, or park the fleet for good.