Sarah Lin: Nathan, I want to start somewhere specific — it's 2 a.m., Houston, 2024, and one person is responsible for 38 autonomous vehicles running across three shifts. One person. And that same program is what Tesla is now pointing to when it says Full Self-Driving Supervised is 5.2 times safer than manual driving.
Dr. Nathan Hayes: The 38-vehicle figure — that's from Medrano's lawsuit, July 28th, 2026?
Sarah Lin: Federal wrongful-termination suit, yes — Javier Medrano, former Tesla self-driving fleet manager in Houston. He alleges the fleet became, um, his exact phrase is 'rolling hazards on public streets.' And he says he was fired for raising those concerns.
Dr. Nathan Hayes: And Tesla Europe releases the safety data the following morning — July 29th. Sixty-five million kilometers across the Netherlands, Lithuania, Estonia, Denmark, Belgium. Zero highway collisions across 41.9 million of those kilometers.
Sarah Lin: So those two things land in the same 24-hour window and — I mean, I don't know how to hold them both at once. The 5.2× claim and the lawsuit. They're describing the same technology.
Dr. Nathan Hayes: Importantly, they're describing different moments of the same technology — that's the methodological crux. The EU data covers April through July 2026. Medrano's account is mid-2024. That is twenty months of distance, and what happened in that gap is genuinely unknown.
Sarah Lin: Right — but does the distance actually close what Medrano is describing, or does it just... stop the documentation of it?
Dr. Nathan Hayes: That is the question this entire episode turns on.
Sarah Lin: But before we can even ask what changed — we have to ask what the 5.2× is actually measuring. Because I don't think it's measuring what most people heard.
Dr. Nathan Hayes: No, it's not. And this is the part that matters. Tesla is comparing FSD Supervised against manually driven Teslas that already have Active Safety features running. Not average drivers in average cars. Their own drivers, in their own vehicles, with a suite of collision-avoidance tools already active. That is the baseline.
Sarah Lin: Wait — so the comparison isn't FSD versus a normal person driving a normal car.
Dr. Nathan Hayes: Correct. And the analogy I'd use — think of a restaurant grading its own health inspection and posting five stars on the window. That's not fraud, it's just... not independent. Tesla gathered the data, Tesla labeled the data, Tesla released the data. No external verification. And this is the part that's actually hard to dismiss as just methodological nitpicking. Reuters published an investigation, roughly May or June 2026, alleging that Tesla's FSD safety statistics are inflated by up to 3×. Three times. Sourced from nine of Tesla's own data labelers, a former self-driving engineer, and eleven independent traffic-safety researchers.
Sarah Lin: The people who actually train FSD.
Dr. Nathan Hayes: Yes. And those same labelers told Reuters they do not trust the technology to drive them personally. That's — I mean, that's the people building it.
Sarah Lin: So did Tesla address any of that before releasing the EU data? Like, three months passed between the Reuters piece and July 29th.
Dr. Nathan Hayes: They released the same category of self-reported, internally verified data as if the Reuters investigation never ran. No methodology response, no independent audit attached. The EU release doesn't acknowledge the critique at all.
Sarah Lin: That silence — um, that's its own kind of answer, isn't it.
Dr. Nathan Hayes: That silence does answer something — but here's the take I keep seeing that I want to push on. People are saying: Houston was 2024, Europe is 2026, these are separate programs. And I understand why that framing is appealing. Twenty months is real time. Conditions can change. But the technology in both is Full Self-Driving Supervised — same system. The Robotaxi that launched in Houston and Dallas in April 2026? Built directly on what Medrano's program produced.
Sarah Lin: And the EU data collection started April 10th, 2026.
Dr. Nathan Hayes: Same month. Which means — now, this is where I'd push the 'separate programs' take hardest — we don't know what, if anything, changed operationally between Medrano's 2 a.m. shift managing 38 vehicles alone and the moment Tesla started collecting the data it handed to European regulators. Tesla hasn't addressed that gap publicly. Not once.
Sarah Lin: Wait — they haven't said anything about what changed?
Dr. Nathan Hayes: Nothing. And that's the actual structural problem. Not that the EU numbers are false — I can't say that. But if the operational foundation was genuinely compromised, and the system built on that foundation is what gets deployed, and then the deployment data is what gets reported... you can't just skip over whether the foundation mattered.
Sarah Lin: Mm. Because imagine — um, picture that 2 a.m. alert. One of Medrano's 38 vehicles flags an incident in a school zone. He's the only manager on shift. If a second alert comes in simultaneously, he cannot physically respond to both. That's not a staffing footnote. That is the operational reality underneath the data that trained this system.
Dr. Nathan Hayes: And the thing that doesn't fit the 'separate programs' story — the Tesla Robotaxi didn't emerge from some clean second chapter. It came out of Houston. That program. The continuity is direct.
Sarah Lin: So when someone says 'maybe conditions improved in twenty months' — maybe. I'll hold that. But Tesla hasn't said what improved, or when, or how. That's a maybe with no floor under it.
Dr. Nathan Hayes: Right — and that lands us at a question we haven't gotten to yet: European regulators are now holding this self-reported data with no independent audit mechanism, and a U.S. senator has already flagged NHTSA. What either of them actually does next — that's where this either becomes consequential or just disappears.
Sarah Lin: And that's the part that actually matters — European regulators are the same bodies Tesla allegedly presented misleading data to before this release, per the Reuters investigation. And now they're holding the July 29th numbers with, as far as anyone can tell, no independent audit mechanism at all.
Dr. Nathan Hayes: No disclosed mechanism. Netherlands, Lithuania, Estonia, Denmark, Belgium — none of them have publicly named an auditing process for Tesla's telemetry or their collision-counting methodology. They're expanding FSD Supervised access on data they cannot independently verify.
Sarah Lin: That's — wait, so the expansion is already proceeding?
Dr. Nathan Hayes: The data was released to support ongoing operations. And here's the — actually, what makes this harder to resolve is there's a third concurrent thread. TechCrunch reports FSD is under a separate investigation for traffic safety violations. Not the lawsuit, not the Reuters methodology critique — a distinct line of regulatory scrutiny. So you now have three simultaneous challenges to the same system's credibility.
Sarah Lin: Three. At the same time.
Dr. Nathan Hayes: And on the U.S. side — a senator has already called for NHTSA to scrutinize Tesla's FSD safety data directly. Which means the credibility question isn't only in Brussels or Amsterdam, it's also sitting with the federal agency that would have to act if something goes wrong domestically.
Sarah Lin: So imagine — um, a road safety official in the Netherlands, right now, trying to decide whether to expand FSD Supervised beyond the current approved corridors. She's reading Tesla's 5.2× figure. She has no audit tool. And somewhere in a file she hasn't opened yet is the Reuters piece saying Elon Musk's safety claims about FSD may be inflated by up to 3×. What does she actually do with that?
Dr. Nathan Hayes: That's — yeah, that's the concrete consequence. She either demands independent verification, which no regulator has publicly required yet, or she makes an expansion decision on trust. And the central unresolved question — what evidence would actually settle this — no one has named that standard. Not Tesla, not NHTSA, not the EU bodies. There is no publicly stated threshold.
Sarah Lin: And without that threshold, the 5.2× number just... stays. Unchallenged. Not because it survived scrutiny — because the scrutiny doesn't have a form yet.
Dr. Nathan Hayes: And that might be the first question, actually. Not 'are these numbers real' — but 'show us the methodology you used to count a collision.' Because what Reuters found, what those nine data labelers described, is that the definition of what gets logged as an incident is — now, Tesla controls that definition. A regulator walking in tomorrow morning needs to ask: what is your collision taxonomy, and who built it.
Sarah Lin: Mm. And what would the answer have to look like to be credible? I mean — um, not just 'here's our methodology document.' Because Tesla already has those. It's something more like... independent access to the raw telemetry? Someone outside the company running the same count?
Dr. Nathan Hayes: An auditor who wasn't hired by Tesla. That's the threshold. And no one — not in the Netherlands, not NHTSA — has named that standard out loud yet. So Medrano's lawsuit is on the public record as of July 28th, the Reuters critique is unanswered, and five EU countries are operating on data that has never been touched by an independent hand. That's just... where it sits.
Sarah Lin: The question I can't put down is whether Medrano ever gets an answer. Not legally — I know that's a separate track. But whether what he described at 2 a.m. in Houston ever becomes part of the official record that shapes how this technology gets judged.
Dr. Nathan Hayes: I don't know. I genuinely don't. Thank you for pushing on the parts I kept trying to make clean.
