Tesla marked a year of Full Self-Driving (Supervised) in Australia and New Zealand on 18 September 2026 by publishing what the fleet did over that year. The window runs from 17 September 2025 to 16 September 2026, and the sample is the largest FSD safety dataset released for any market outside North America.
The numbers
| With FSD (Supervised) | Driven manually | |
|---|---|---|
| Distance covered | 160 million km | ~2.4 billion km |
| Collision rate | 1 per 1.78 million km | 1 per 1.07 million km |
That works out at roughly 40% fewer collisions per kilometre. Behind the ratio sit about 90 collisions on FSD and something over 2,200 without it — counts large enough that one more crash either way changes nothing.
The secondary measures move further than the headline:
| Measure | Change with FSD engaged |
|---|---|
| Automatic emergency braking events | 66% fewer |
| Harsh braking | 90% fewer |
| Harsh acceleration | 86% fewer |
| Harsh cornering | 76% fewer |
| Horn use | 80% fewer |
The horn figure has the clearest units attached: once every 3,233 km with FSD engaged, against once every 643 km under manual control.
Why the sample size is the story
Tesla published Belgian-style claims all year, but the evidence has mostly been thin. Flanders gave its parliament its own FSD crash figures on the same day, and that dataset — genuinely independent, gathered by a member-state authority — rests on a total of five collisions. One additional crash would move its ratio a long way.
The Australian and New Zealand data is the mirror image: hundreds of times more incidents, so the rate is stable, but Tesla collected it, published it, and is the only party able to check it. Neither dataset is the one Europe actually needs. That would be a regulator's audit of a fleet this size.
The caveats Tesla states, and the one it doesn't
Tesla labels the figures approximate, drawn from all road types, and combines both countries — so there is no Australian number and no New Zealand number, and no per-road-type split of the kind Flanders provided. Tesla has not said how many cars are in the dataset or how many hold an FSD subscription.
The structural caveat is the same one that attaches to every fleet comparison of this kind. Drivers choose when to switch FSD on, and they choose the roads and conditions that suit it; the manual baseline includes everything else, including the situations nobody would hand over. And because the system is supervised, what is being measured is a driver and the software together. A human who takes over and prevents a crash appears in these figures as FSD not crashing.
What it means for Europe
Australia and New Zealand were the fifth and sixth markets to get FSD (Supervised), in September 2025, and the pair have since become Tesla's clearest demonstration case outside the US — most recently when Zurich began pricing FSD into Australian premiums on 14 September, only the second insurer worldwide to do so. An insurer acting on the data is a stronger signal than the data alone: Zurich is putting its own money behind the claim that these cars crash less.
The timing is not incidental. The European Commission's Technical Committee on Motor Vehicles votes on 6 October on extending FSD approval, and the evidence in front of it is overwhelmingly Tesla's own. A year of Australian and New Zealand fleet data arriving three weeks before that vote is a submission to it in all but name — which is a reason to read it carefully rather than a reason to discount it.