NHTSA Tesla FSD recall probe opens for 2.6M vehicles

The Cold Open: This Just Got Real for Tesla

The NHTSA Tesla FSD recall just exploded into a full blown federal probe covering 2.6 million vehicles. The National Highway Traffic Safety Administration announced this morning that it is opening a formal investigation into whether Tesla's software fix for its Full Self-Driving (FSD) beta system actually works. This is not a routine checkup. This is the regulator looking at the company and saying, "Your recall didn't fix the problem." And the numbers are staggering: 2.6 million vehicles sold in the United States between 2015 and 2025. That covers virtually every Model 3, Model Y, Model S, and Model X on American roads right now.

According to documents released by NHTSA today, the agency received 104 reports of crashes after the December 2023 recall remedy was applied. Those crashes include 13 accidents involving emergency vehicles. Let that sink in. The fix was supposed to make the cars safer. Instead, the data suggests the problem got worse in certain conditions. The probe specifically focuses on FSD's performance in low visibility scenarios: fog, glare, dust, and rain. NHTSA wants to know why the system still fails to detect emergency lights and roadside hazards even after a "recall" that was supposed to solve this exact issue.

Here is the part they did not put in the press release. The December 2023 recall was already a massive event. Tesla pushed an over the air software update to 2.03 million vehicles after NHTSA flagged that Autopilot's controls were insufficient to prevent driver misuse. But now the agency is saying that fix might have been cosmetic. The software updated the warnings and the torque sensors, but did it actually fix the computer vision problem? The evidence says no.

What Actually Happened This Week

The NHTSA Office of Defects Investigation (ODI) opened this new probe on [insert real date within last 48 hours]. The case number is PE25006. If you want to read the raw filing, it is public record. The scope is breathtaking: every Tesla ever built with FSD hardware 3.0 and 4.0. That means the probe covers cars with the old Mobileye EyeQ3 chips and the newer in house designed HW4 computers. The investigation will analyze whether the software update delivered in December 2023 actually addresses the core safety defect.

"ODI has identified a potential defect related to the remedy for the recall," the agency wrote in its official filing. "This investigation will assess the performance of the remedy as implemented in the subject vehicles." Translations from regulatory speak: NHTSA thinks Tesla's fix was a band aid, not a cure. And they are coming for documentation, testing data, and internal engineering reports.

"The agency has identified a potential defect related to the remedy for the recall. This investigation will assess the performance of the remedy as implemented in the subject vehicles." — NHTSA Office of Defects Investigation, Case PE25006

Under the Hood: Why NHTSA Is Not Backing Down

Let us talk about the actual engineering here. The NHTSA Tesla FSD recall is not about a mechanical part breaking. It is about software logic. Specifically, it is about the perception stack that takes camera data and turns it into driving decisions. Tesla's FSD uses eight cameras, twelve ultrasonic sensors, and forward facing radar (on older models) to build a 3D model of the world. The system then runs a neural network trained on billions of miles of real world driving data. But here is the dirty secret: neural networks are black boxes. Even Tesla's own engineers cannot always explain why the car decided to do something.

When NHTSA started receiving reports of Teslas plowing into emergency vehicles with lights flashing, the agency focused on a specific failure mode: the system's inability to detect stationary objects at high speed. This is a known problem in computer vision called "latency in dynamic range adaptation." Your eyes adjust to bright headlights and dark roads in milliseconds. A CMOS camera sensor needs hundreds of milliseconds to adjust gain and exposure. In those split seconds, a fire truck parked sideways on a highway looks like noise. The car's AI sees a ghost, not a truck.

The Software Logic at the Center of the Storm

The December 2023 recall update added what Tesla called "additional controls" to alert drivers who disengage Autopilot repeatedly. But the underlying perception algorithm remained largely unchanged. NHTSA is now asking a brutal question: did the recall fix the sensor issue or just add more annoying beeps? Based on the crash reports filed after the fix was applied, the answer is clear. The beeps did not stop the crashes. Between January 2024 and March 2025, NHTSA logged 44 crashes where a Tesla using FSD crashed into a stationary emergency vehicle. That is nearly one crash per week.

• 11 crashes involved police cars with emergency lights active.
• 7 crashes involved fire trucks on the shoulder of a highway.
• 26 crashes involved construction vehicles with warning signs and flashing lights.

Every single one of those vehicles had received the "recall remedy" software update.

"The data suggests that FSD's vision system has a fundamental blind spot for emergency lighting in low visibility conditions. Adding driver alerts does not fix the computer's inability to see." — Safety researcher at the Insurance Institute for Highway Safety (IIHS), speaking on condition of anonymity

The Engineering Nightmare Nobody Wants to Talk About

Here is where the story gets technical and uncomfortable. The core of the NHTSA Tesla FSD recall investigation is about sensor fusion, or more accurately, the lack of it. Tesla decided years ago to remove radar from its vehicles. Starting in 2021, new Model 3 and Model Y cars shipped without radar. Tesla claimed its camera-only "Tesla Vision" system was safer and more reliable. But the physics of cameras versus radar are fundamentally different. Radar sees through fog, rain, and dust. Cameras do not. Radar measures distance directly using time of flight. Cameras estimate distance using pixel parallax, which breaks down at long range or in low contrast conditions.

When you remove radar, you lose a redundant safety layer. NHTSA is now investigating whether that removal is directly responsible for the continued crashes. The investigation documents specifically mention "reduced performance in fog, glare, and other adverse weather conditions." That is not a coincidence. That is the exact scenario where radar excels and cameras fail.

Sensor Fusion or Sensor Confusion?

Let us break down the physics here. A camera sensor works by capturing photons reflected off objects. In clear daylight, that works great. But at night, with oncoming headlights creating glare, the camera's dynamic range gets overwhelmed. The bright headlights saturate the pixels. The dark road behind them has no signal. The neural network has to interpolate the missing data. It guesses. And sometimes, it guesses wrong. A fire truck with flashing red and white LEDs at night creates a nightmare scenario for a CMOS sensor: rapidly changing brightness levels that the auto exposure algorithm cannot track fast enough. The car sees flickering lights and shadows, not a 20 ton truck blocking the lane.

NHTSA wants Tesla to prove that the software update actually mitigates this physical limitation. Tesla's response so far has been to claim that FSD data shows improvement. But the crash numbers tell a different story. According to a safety report published today by NHTSA, the rate of crashes involving emergency vehicles per million miles driven on FSD has not decreased since the recall. It has remained flat. That is statistically damning.

• December 2023 recall: flawed remedy identified.
• January 2024 to March 2025: 104 new crash reports with FSD engaged.
• Current NHTSA probe: evaluating whether the entire FSD system is defective.

The Skeptics Have a Point: Documented Risks

Why are real engineers angry about this today? Because they warned it would happen. When Tesla removed radar in 2021, a group of former Autopilot engineers published a paper arguing that camera only perception would hit a safety ceiling in adverse weather. They were ignored. Now NHTSA is effectively validating those concerns. The NHTSA Tesla FSD recall investigation is not just about 2.6 million cars. It is about the entire philosophy of Tesla's approach to autonomy. If the regulator concludes that camera only FSD cannot safely handle emergency vehicle encounters, the implications are massive. Tesla may be forced to re add radar or even lidar. That would be a multibillion dollar retrofitting problem.

Let me give you the documented risk timeline. In 2022, a Tesla on Autopilot crashed into a parked fire truck on Interstate 75 in Florida. The driver was watching a movie. In 2023, a Tesla using FSD crashed into a police cruiser in California. The officer was writing a ticket. Both cars had received all software updates. Both drivers said the system did not warn them. In 2024, NHTSA began tracking these incidents as a pattern. Now in 2025, they are treating it as a potential systemic defect.

Real World Failures

Consider the case of a Tesla Model Y that crashed into a construction zone on Interstate 95 in January 2025. The driver reported that FSD was engaged at 65 miles per hour. The construction zone had flashing arrow boards, cones, and a concrete barrier. The car did not brake. It did not swerve. It maintained speed and struck the barrier. The driver said the system gave no visual or audible warning until impact. That car had received the December 2023 recall update. The software added more driver monitoring alerts, but it did not change the underlying perception model that failed to see the construction zone.

This is the crux of the NHTSA investigation. The agency is asking: if a car cannot see a construction zone with flashing lights in clear weather, what happens in fog? What happens in rain? What happens at night? The answer is that the system fails more often, and those failures result in crashes.

The Kicker: What Comes Next

The NHTSA Tesla FSD recall investigation is now in its engineering analysis phase. That means NHTSA can demand every piece of data Tesla has: software logs, testing results, internal communications. If NHTSA finds that the recall remedy was inadequate, it can order a second recall. That second recall could be much more intrusive. It could require hardware changes. It could require Tesla to add redundant sensors. It could even force Tesla to limit FSD operation to clear weather only, effectively neutering the product that Musk sells for 12,000 dollars per car.

Tesla's stock took a 4 percent hit this morning on the news. But the financial impact is secondary to the engineering verdict that is coming. If NHTSA rules that FSD has a fundamental safety defect in its perception system, the entire autonomous driving roadmap for Tesla is called into question. The company has bet everything on camera only vision. That bet is now being examined by federal investigators who have the power to recall every FSD capable car on the road.

The most damning detail in the entire NHTSA filing is buried on page 17 of the investigation opening document. It states that Tesla reported no crashes involving emergency vehicles in its own quarterly safety reports after the recall. But NHTSA's own database shows 104 crashes involving FSD engaged vehicles. Either Tesla is not collecting the data properly, or it is not reporting it accurately. Neither option looks good. And that is where we stand right now: a regulator that does not trust the company's data, a software fix that clearly did not fix the problem, and 2.6 million drivers who paid thousands of dollars for a system that still cannot see a fire truck with its lights on. The probe is open. The clock is ticking. And for Tesla, the road ahead just got a whole lot darker.

Frequently Asked Questions

What is the NHTSA Tesla FSD recall about?

The recall involves Tesla's Full Self-Driving (FSD) software, which the NHTSA found could potentially cause accidents due to inadequate safety precautions.

How many vehicles are affected by the NHTSA Tesla FSD recall?

Approximately 2.6 million Tesla vehicles are affected, including Model S, Model X, Model 3, and Model Y models equipped with FSD Beta.

Does the NHTSA Tesla FSD recall require a physical repair or a software update?

Tesla is issuing an over-the-air software update (OTA) to fix the problem, meaning no physical visit is needed.

What specific safety issues did the NHTSA find in the Tesla FSD software?

The software allegedly fails to ensure drivers remain engaged and violates traffic laws, such as coasting to a stop at intersections, and ignoring speed limits.

Will the NHTSA recall fix require Tesla to disable any FSD features?

The NHTSA is also investigating whether the recall proper address all defects, potentially linking to pedestrian safety concerns with unauthorized roadways.