Tesla redacted 3,951 of its 3,953 federal crash reports

Tesla has filed 3,953 crash reports under NHTSA's Standing General Order for driver-assist systems. Journalist Fred Lambert found that all but two are stamped [REDACTED — confidential business information] — including reports on crashes that killed people.

Two

That is how many of Tesla's federal crash filings are readable.

Under NHTSA's 2021 Standing General Order, automakers must report any crash where a Level 2 driver-assist system was engaged within 30 seconds of impact and someone was injured, someone was killed, or a vehicle was towed. Tesla accounts for roughly 85% of the entire industry's reports — nearly 4,000 of them — and the count sets new records every year.

Of 3,953 Tesla filings, Fred Lambert found that only two do not carry the label [REDACTED — confidential business information], according to a summary of his research. On 3 September 2026 he added a fifth fatal crash to an ongoing effort to match those blacked-out filings to publicly reported incidents: an Ohio crash in which a motorcyclist was killed after being struck by a Model X with an ADAS system active.

The redacted fields are not incidental. They are the ones that matter: what the car did, which software was running, and whether the car was even supposed to be operating that way. Tesla's stated justification is that it would suffer financial harm if the information became public. Electrek's read on that is correct and worth repeating without softening: the company is telling a federal regulator that the commercial value of its secrecy outweighs the public's ability to learn how its cars behave in fatal crashes.

What the black bars are hiding

The most recent data from Tesla and NHTSA show 207 crashes involving Teslas in Autopilot or self-driving modes in May — the most in any month ever, per reporting relayed by NewsNation. Comparing the first half of 2026 with previous years, the brand is on track to surpass its earlier crash totals for vehicles in Autopilot mode. And the outlet notes the same thing Lambert did: Tesla redacted the information that would identify whether the cars were in Autopilot or Full Self-Driving at the time.

So the public number goes up, and the field that would let anyone attribute it goes dark. That is a choice, not a technical limitation.

The caveat that cuts our way and against us

Here is the part a lazy version of this article would leave out. Electrek says it plainly: between FSD and Autopilot, Tesla almost certainly has more ADAS mileage than any other automaker, so having more crash reports than everyone else is not necessarily a problem. Tesla had 1.48 million active FSD subscriptions as of August, growing 56% year over year. A bigger denominator produces a bigger numerator. Raw report counts prove very little on their own.

Which is exactly why the redactions matter so much. The denominator argument is Tesla's best defence, and Tesla is the only party with the data to test it. It has chosen to withhold that data while continuing to make public safety claims that depend on it.

Meanwhile, in Brussels

On 1 September, ahead of a possible EU-wide approval vote as soon as 6 October, Tesla published a safety dashboard claiming FSD is 4.1 times less likely to be involved in a crash than manual driving across 100 million km of EU roads: three highway collisions and nine on non-highway roads, against 137 and 490 respectively for manually driven vehicles.

Now watch the trend. On 28 July the company claimed FSD was over 5.2 times safer across 65 million km from 10 April to 26 July, with zero collisions in 41.9 million highway km. The Netherlands-only figure in June was 3.5 times. Between the July and September releases, mileage grew 54% while the multiple fell 21%. The 4.1x figure comes from the customer fleet after April and appears nowhere in the underlying file.

And the file itself? Automotive World reports its newest data is nine months old — used to support a safety claim about a rollout that only began in April 2026. Tesla's own methodological notes concede the fleet-wide comparisons do not control for traffic density, weather, road surface, time of day or scenario complexity, and describe the results as "observational associations under stratified exposure conditions, not as causal estimates from a randomised experiment." The company's public line is that "FSD is already saving lives where it's approved." Its own filing does not support that sentence. Reuters has previously reported that Tesla presented European regulators with safety statistics experts said relied on invalid comparisons.

Regulators are not entirely buying it. The European Transport Safety Council wrote to transport ministers on 12 June urging them not to recognise the Dutch approval until there is meaningful, published real-world safety evidence. France's transport minister said on 22 July the system would not be authorised in its current form, citing speeding and driver-attention concerns; on 3 September, after a meeting with Musk, France authorised road tests — two vehicles, drivers must remain attentive, no public use.

Independent eyes

An AMCI Testing evaluation over more than 1,000 miles in fair-weather conditions found good vehicle control and lane positioning alongside persistent route-planning mistakes, occasional glitches, driver complacency and unvalidated severe-weather capability. Investor Ross Gerber reported on 31 July that FSD twice slammed on the brakes as emergency vehicles passed, saying the technology "still has major issues". Waymo, after more than 200 million fully autonomous miles, published ten lessons whose sharpest line is that turning a driver-assist system into full autonomy is a "false summit". Meanwhile Tesla owners are filming themselves defeating cabin-camera monitoring — a folded $20 bill over the lens — and posting it for clout.

Why we're the ones saying it

This site exists because the fear attached to electric cars is manufactured while the harm from burning fuel is ignored. That harm is measured: fine particulate matter, ozone and nitrogen dioxide were responsible for an estimated 74,000 premature deaths in Europe in 2024, and lifecycle greenhouse gas emissions for a battery car are 73% lower than a gasoline car. None of that changes.

But that case does not need Tesla's black bars, and it is weakened by them. Every redacted field hands ammunition to people who argue the whole transition runs on unverifiable claims. Tesla is asking Europe to approve a system on the strength of numbers only Tesla can check, in the same month it emerged that the company has redacted 3,951 of its 3,953 US crash reports.

Publish the fields. Two out of 3,953 is not transparency. It's a press release with a database attached.

Straight answers

How many of Tesla's NHTSA crash reports are redacted?

Of 3,953 Tesla crash reports filed under NHTSA's Standing General Order for Level 2 driver-assist systems, only two do not carry the label "[REDACTED — confidential business information]," according to research by journalist Fred Lambert published in September 2026. The redacted fields include what the vehicle did, which software version was running, and whether the vehicle was operating as intended.

What share of federal ADAS crash reports come from Tesla?

Tesla accounts for roughly 85% of all crash reports filed under NHTSA's 2021 Standing General Order — nearly 4,000 reports. Electrek notes this is not automatically damning, because Tesla likely has more driver-assist mileage than any other automaker, so a larger denominator produces more reports. The problem is that Tesla redacts the data needed to test that defence.

Is Tesla's claim that FSD is 4.1 times safer than manual driving reliable?

Tesla's 1 September 2026 dashboard claims FSD is 4.1 times less likely to be involved in a crash across 100 million km of EU driving. Tesla's own methodological notes state the comparisons do not control for traffic density, weather, road surface, time of day or scenario complexity, and describe the results as "observational associations under stratified exposure conditions, not as causal estimates." Automotive World reports the newest data in the underlying file is nine months old.

Why has Tesla's own safety multiple been falling?

Tesla claimed over 5.2 times fewer collisions across 65 million km on 28 July 2026, then 4.1 times across 100 million km on 1 September. Mileage grew 54% between the two releases while the claimed safety multiple fell 21%. The Netherlands-only figure in June 2026 was 3.5 times.

Has any European country blocked Tesla FSD approval?

France's transport minister said on 22 July 2026 that FSD would not be authorised in its current form, citing speeding and driver-attention concerns, and France opposes EU-wide approval as it stands. On 3 September France authorised limited road tests using two vehicles, with attentive drivers and no public use. The Netherlands granted provisional approval on 10 April 2026, followed by Belgium, Denmark, Estonia and Lithuania. An EU-wide vote is expected as soon as 6 October 2026.