NHTSA Goes Looking For The Radar Tesla Threw Away
There’s a genre of internal engineering document that no company ever wants a federal investigator to see, and it’s the one with a blunt title. Somewhere inside Tesla is a file called “Radar Saves Us.” The Office of Defects Investigation knows it exists because Tesla handed it over, and in the information request letter it sent Tesla on July 2, ODI asked for it again — by name, along with every email, test record, and dataset behind it.
That request lands inside EA26002, the Engineering Analysis NHTSA opened on March 18 after upgrading its earlier Preliminary Evaluation into Full Self-Driving crashes in reduced visibility. An EA is the last rung on the ladder. From here the agency either closes the file or starts pushing for a recall, and the population in scope is 3,203,754 vehicles: 2016–2026 Model S and X, 2017–2026 Model 3, 2020–2026 Model Y, and 2023–2026 Cybertruck. Tesla’s answers are due August 12.
Related Articles
- Mazda and Hyundai Just Turned China Into Their EV Shortcut
- The 10 Most Insanely Expensive Headlights on Sale Today (One Pair Costs More Than a New Porsche)
Read alongside the letter, ODI’s stated interest is narrow and pointed. It wants the document because it bears on design changes to the degradation detection system and on what Tesla understood about the “limitations imposed by a vision-only OEDR system.” OEDR — object and event detection and response — is the SAE J3016 term for the part of the driving task that involves noticing things and reacting to them. Translated out of regulator-speak: NHTSA thinks Tesla has already written down, internally, what happens when the only sensor you have stops working.
What’s already in the file
You don’t need the July letter to see how bad the underlying record looks, because the March opening resume is public and it is not kind.
Nine crashes qualify, one fatal, two with injuries. Six more are under review. The pattern repeats: sun glare, fog, or airborne dust degrades the cameras, the degradation detection system either doesn’t flag it or flags it too late for a human to do anything useful, and FSD either loses the lead vehicle or never sees it at all. Tesla’s own post-incident analysis told ODI that its updated degradation detection system, had it been installed, might have changed the outcome in three of the nine.
The timeline is the part that should make a product liability attorney sit up. Tesla filed its Standing General Order report on the November 28, 2023 fatal crash on June 27, 2024 — roughly seven months late. It began developing the degradation detection update on June 28, 2024. One day later.
Tesla also told investigators that its internal data labeling made it hard to consistently identify crashes with FSD engaged, which ODI reads as a plausible source of underreporting across the review period. When a manufacturer’s crash-counting method is itself in question, every safety statistic built on top of it gets softer.
The marketing thread isn’t new either
The July letter reportedly asks Tesla to go statement by statement through its own promotional claims and say whether each implies more capability than the system has. That’s an escalation, not a new idea. It was already an enclosure in the November 2024 opening letter, where ODI attached a May 2024 email cataloguing Tesla’s reposts: the owner who let FSD drive him to the ER during a heart attack, the post-stadium drive captioned “$99 / month,” the hands-off video Tesla amplified with a note about reduced driver fatigue. The agency also flagged Tesla’s own website, including the Autopilot page video that opens by telling you the person behind the wheel is there for legal reasons only.
NHTSA asked Tesla in 2024 to make its public messaging match what it tells regulators. Whatever happened next, the agency is now asking Tesla to certify the gap in writing, under an information request backed by 49 U.S.C. § 30166. That matters because § 30166 responses carry penalties of $27,874 per day and up to $139,356,994 for a related series — and because sworn characterizations of your own advertising become discoverable in every civil suit that follows.
The “no such radar existed” defense has an expiration date
Tesla removed radar from Model 3 and Model Y in 2021, from Model S and X in 2022, then deleted ultrasonic sensors in 2022 and 2023. The company’s public position was that vision plus neural net processing matched or beat radar-equipped cars on active safety ratings.
Related Articles
- Carvana Didn’t Beat the Franchise Dealers. It Just Became One — And Stellantis Is Footing the Bill
- The Grappler Was Built To End Deadly Police Chases. Almost Nobody Actually Has One
Here’s the wrinkle that rarely makes it into the argument: Tesla certified its own high-resolution radar with the FCC in June 2022. The test report for FCC ID 2AEIM-1541584 describes “a non-pulsed Automotive Radar which operates in the 76-77 GHz spectrum” with three sensing modes. That unit — known internally as Phoenix — began appearing behind the front fascia of Hardware 4 Model S and Model X builds from 2023, mounted near the passenger-side headlight with its own heater.
So a chunk of the fleet already carries a 76–77 GHz imaging radar that has spent years doing, as far as anyone outside Tesla can tell, nothing. If “Radar Saves Us” turns out to be the analysis that justified buying it, ODI will have a document showing the company identified a sensing gap, engineered a fix, certified the hardware, bolted it to some cars, and left the software switched off across 3.2 million vehicles. Knowledge plus an available remedy plus no action is the exact shape of an unreasonable risk finding under 49 U.S.C. § 30118.
What this means if you own one
Practical stuff, because the regulatory outcome is months away and your Tuesday commute isn’t.
FSD Supervised is SAE Level 2. Legally, you are the driver in every state, in every condition, including the ones the cameras can’t handle. Tesla’s own owner’s manual lists poor visibility from heavy rain, snow, and fog among the conditions that degrade Autosteer, along with bright light “interfering with the view of the camera(s).” Your insurer will read that manual before it reads your account of the crash, and no carrier currently discounts for FSD.
Camera health is now a maintenance item, not a detail. On camera-only cars the windshield is a sensor housing. A cheap aftermarket glass replacement with the wrong optical clarity, a badly aimed camera bracket, or a skipped calibration will quietly degrade the same perception stack this investigation is about. Insist on calibration documentation after any windshield or bumper work, and treat repeated “camera obstructed” or degraded-vision alerts as a service visit rather than an annoyance.
Know which hardware you bought. HW3 cars cannot be made into HW4 cars with a software update, and hardware generation is becoming the dividing line for resale value on FSD-equipped Teslas. If a used listing claims FSD capability, check the actual Autopilot computer, not the window sticker.
Whether Tesla answers on time
Worth watching. In a January 2026 extension request on a separate FSD probe, Tesla told ODI it was juggling three large information requests at once, could process roughly 300 records a day, and that the pileup was “unduly burdensome and affects the quality of responses.” It got five extra weeks.
There’s also a decent chance the public never sees the answer. Tesla routinely claims confidential treatment under 49 C.F.R. Part 512 — including for its July 2025 presentation on Direct Photon Count, the camera improvement it has pitched as a glare fix — and at least one prior response in this same investigation was withheld in its entirety because the confidentiality claim covered every page.
Which is the quiet lesson here. The most consequential document in this file has a title everyone can read and contents almost nobody will.