What the federal crash reports actually say about Tesla's robotaxis

Arguments about robotaxi safety usually run on vibes: a viral clip of a car doing something stupid, or a company statistic with no context. There is a better source. Companies operating autonomous vehicles must report crashes to the National Highway Traffic Safety Administration, and those filings are public.

Here is what Tesla's filings show, as reported by Electrek and TechCrunch.

The record: 17 incidents, mostly minor

Across July 2025 through March 2026, the whole period of robotaxi testing in Austin, Tesla reported 17 unique incidents involving its autonomous driving system:

Outcome	Count
Property damage only	13
No injuries reported	2
Minor injury, no hospitalization	1
Minor injury requiring hospitalization	1

No deaths, no catastrophic crashes. For a fleet learning to drive itself in a real city, that is a meaningful result, and it deserves to be said plainly.

Who was at fault

Only six of the incidents were minor contact events where the Tesla system was judged at fault or partly at fault. A notable share of the rest involved the Tesla being rear-ended while completely stopped, at red lights, at stop signs and in traffic.

That pattern is familiar from other autonomous fleets, and it points at something real: self-driving cars follow the rules more literally than human drivers expect. A car that stops fully and waits is a car that human drivers sometimes run into.

The uncomfortable detail: the humans watching

Two of the crashes did not involve the software at all. They involved Tesla's remote operators, the staff who take over when the car gets stuck.

• In July 2025, a teleoperator drove a robotaxi up a curb and into a metal fence at about 8 mph after the automated system could not resolve a situation. That incident produced the one injury in the dataset that required hospitalization.
• In a separate case reported in July 2026, a remote operator drove a robotaxi into a tree stump in Houston, according to Electrek's review of NHTSA data.

This matters more than it might seem. The safety debate is usually framed as computer versus human driver. Driverless fleets today are actually a hybrid: software drives, and humans intervene remotely. The handoff itself is a risk surface, and it is the part almost nobody is measuring.

The wider federal picture

Tesla's autonomy work sits under broader scrutiny. NHTSA has an open investigation into Full Self-Driving covering roughly 3.2 million vehicles after visibility-related crashes, and, as of September 3, the separate audit query into how Tesla certified the Cybercab.

How to read any robotaxi safety claim

1. Ask for miles, not incidents. Crashes without miles driven tell you nothing.
2. Ask who was at fault. Being hit while stopped is not the same as causing a collision.
3. Ask about the humans. Remote operators and safety monitors are part of the system.
4. Ask about conditions. Fair-weather daytime miles in one city are not the same as rain at night across a region.
5. Watch the trend, not the single clip that went viral.

On the current record, Tesla's robotaxi program has not produced a safety catastrophe. It also has not yet produced the volume of miles that would let anyone claim it is safer than a human driver. Both things are true, and the next year of filings will say more than any launch event.

Related: Tesla vs Waymo: scale versus speed

Sources: Electrek and TechCrunch on Tesla's NHTSA crash narratives; Electrek on the Houston teleoperator crash; NHTSA investigation records. Market information and analysis, not investment advice.