Picture this: you're in a Model Y on the highway, and instead of the usual last-second scramble across 3 lanes to catch your exit, the car starts moving over 2.5 miles early. Smooth. Calm. Almost smug about it. One early tester called it "sublime," and honestly, I get it. That feeling is real.
The feeling is also doing a lot of heavy lifting right now.
Tesla rolled out FSD (Supervised) v14.3.1 this week, and the changelog is legitimately impressive. Reaction times are 20% faster, thanks to a rewrite of the AI compiler. The Dutch approved it after 1.6 million km of testing. Elon Musk called v14.3 "the last big piece of the self-driving puzzle," which, sure, he says something like that every 18 months, but the underlying improvements are not fake.
Here's my problem. A YouTuber tested v14.3.1 in mid-April and caught it speeding through a school zone. Not a close call. Not a sensor hiccup. The system just... did not care about the 30 mph limit. The tester's note: "They want you to be at 45 by the time you reach the 45 sign." Kids walk in school zones. That is not a vibe issue.
The Number Tesla Won't Show You
Tesla has fleet data on millions of miles. They know their crash rate. They have not published a comparison to human drivers in any format that an independent researcher can verify. Waymo does this. Tesla does not. Instead, we get reaction time percentages and user testimonials about sublime lane changes.
I'll grant the skeptics one thing: the "no safety-critical takeovers in years" claim from some testers is genuinely interesting data. If true across a large sample, that matters. But "some testers haven't had to grab the wheel" is not a safety record. It's an anecdote wearing a lab coat.
The 20% faster reaction time sounds great until you ask: faster than what baseline, in what conditions, verified by whom? Tesla's own compiler team. That's the whole citation chain.
Why This Matters to You Specifically
If you're paying the $99/month FSD subscription, or you bought it outright for $8,000 (I still think about that number at 2am), you are not just a customer. You are a supervised driver with legal liability for everything the car does. The system is called FSD Supervised for a reason. You are the last line of defense against the school zone problem.
The smoothness is seductive. I have been in these cars. The exits feel like the car read your mind. That feeling makes you trust it more than you should, and Tesla knows this. A calmer ride trains you to pay less attention, which is exactly backwards from what "supervised" requires.
The EU joint safety assessment is scheduled for May 2026. Italy already delayed its rollout. Regulators are asking the right questions. Tesla should answer them publicly, not just in approval paperwork that nobody outside the RDW will ever read.
FSD is getting better. The highway exit fix is real. The reaction time improvement is real. But "better" without a public benchmark is just marketing with extra steps. Tesla should publish quarterly crash rate data versus human drivers, full stop. Until then, the smoothness is a feeling, and feelings have killed people in school zones before.
