Anton wants an independent journalistic deep research piece on the safety of Tesla cars. Study the forums of Tesla employees and former employees, the unions, and the other places where people write that Teslas are not the most reliable cars. Look separately at build quality, how the cars behave on the road, and how safe the autonomous driving is, Autopilot and FSD. What the complaints are and what is said from the inside. Then a podcast on Tesla safety and the quality of their autonomous driving. And a second podcast: did Tesla really turn the world car industry upside down, how much did it push manufacturers toward electric vehicles, or was the switch an inevitable trend of the time that would have happened without Tesla.
The recall was done, and this is the fifth brief in a row where it answers. This time with two reports, both dated 30 October, one per podcast.
On safety: DR26-10-30-HUB-02, "Safety of Tesla and driving automation systems (Autopilot, FSD, Robotaxi)". Its first recommendation is the thing the brief most needs: do not reduce the question to a single verdict. Crash protection, reliability, fires and evacuation, supervised automation and driverless services are four different questions with four different answers.
Crash protection is strong, and that is established. The 2025 Model Y took IIHS Top Safety Pick+, Model 3 took Top Safety Pick, Cybertruck TSP+ for builds after April 2025. Euro NCAP 2025 gave Model 3 five stars (90/93/89/87 percent). A crash test does not measure how often you crash.
In-service reliability in Germany is worse, and that is established for that cohort. TÜV Report 2026, cars aged two to three years: the age-group average is 6.5 percent significant defects, Model 3 sits at 13.1 percent and Model Y at 17.3 percent, the worst in the group. TÜV SÜD names brake discs and suspension. Teslas in that sample run over 50,000 km, more than many of their peers, and this is not proof of a higher crash rate and does not transfer automatically to new facelifts or to the US.
On the thing the brief asks for specifically, employee forums and unions, the report is blunt: employee testimony (the anonymous Reuters data labelers, Krupski) and union conflicts (IG Metall, IF Metall) are not evidence that the cars are defective. They are evidence about working conditions and about labelling practice. Treating them as a verdict on the vehicle is the first trap in this investigation.
Autopilot, historically, is established. NTSB found over-trust in automation and weak attention monitoring in the 2016 Williston and 2018 Mountain View crashes. Recall 23V838 (December 2023) covered 2,031,220 cars, with Tesla conceding insufficient "prominence and scope". NHTSA closed EA22002 in April 2024; RQ24009 then examined 20 crashes after the fix, which tests the remedy rather than blaming it.
FSD (Supervised) is still SAE Level 2. Two NHTSA investigations are open: PE25012 (58 incidents, 14 crashes, 23 injuries, population 2,882,566: red lights, wrong-way lane changes) and EA26002 (9 crashes, 1 fatal, population 3,203,754: degraded visibility). ⚠️ Both denominators are cars, not miles, so no crash rate follows. The counterfactual on electrification cannot be quantified either: "X years of acceleration" would be invented precision.
And the claim the podcast will be tempted to repeat: Tesla's "many times safer with FSD" figures are not causally established. Driver self-selection, road mix, vehicle age, weather, incomparable endpoints, no independent VIN-level audit, no intervention data. Reuters on 28 May 2026 reported that 10 of 11 researchers surveyed found the key comparisons problematic. The association exists; the causal magnitude does not. Any "N times safer than a human" formulation is low confidence.
Robotaxi, where the honest numbers are smaller than the headline. Tesla reported about 3.1 million autonomous miles as of 18 October 2026 with zero events in three selected categories. That is not zero collisions: WIRED counted 17 SGO events between July 2025 and March 2026, at least two with remote operators driving. For scale, Waymo has 271.3 million rider-only miles as of June 2026, at 0.67 injury crashes per million miles against a 3.77 human benchmark, and discloses more detail. "Tesla Robotaxi is safer than Waymo" remains an assumption.
⚠️ One thing the 30 October report does not have, because it is newer: Consumer Reports' 2026 brand rankings put Tesla in the top 10 for the first time, ninth on reliability with a score of 50, over a survey of about 380,000 vehicles, with only the Cybertruck below average. And the opposite side of the same coin: Consumer Reports' 2025 used-car study ranked Tesla last of 26 brands. So the premise "people write that Teslas are not the most reliable" is true of used Teslas and increasingly false of new ones, which makes the real question not "are they reliable" but "which ones, and at what age".
On the second podcast: DR26-10-30-HUB-01, "Tesla and global car electrification: accelerator, passenger, or system". Verdict, at high confidence: the interaction hypothesis. Tesla was neither the originator of electrification nor a passenger without influence. It amplified a transition already under way, together with batteries, policy, China and competitors.
Independent trajectories predate the Roadster: California's ZEV mandate from 1990, the GM EV1 (1,117 built), the Prius from 1997, lithium-ion from 1991, BMW's i3 announced 2009, the BYD F3DM 2008, Chinese NEV subsidies from 2009-2010.
The structural scale belongs to China and to the battery cost curve. By 2025 China had sold over 13 million EVs, about 55 percent of its new cars; Chinese OEMs accounted for roughly 60 percent of world EV sales; China produced about 70 percent of EVs and over 80 percent of battery cells. Lithium-ion pack cost fell from $1,415/kWh in 2008 to $139/kWh in 2023, about 90 percent, per DOE.
And the test that settles the "was it inevitable" question: countries with minimal Tesla presence electrified quickly anyway. Vietnam with VinFast at 175,099 EVs in 2025. Thailand, where BYD and MG lead and Model Y is only tenth. Indonesia, where SAIC and BYD hold over 60 percent of cumulative BEV sales. So Tesla was not a necessary local condition. In the US its role was larger, over half of EV sales in 2021.
Tesla's real channels were narrower than the myth: the BEV as a desirable premium product, a big battery with power and its own Supercharger network, early OTA design, competitive pressure, Model 3 with 325,000 reservations in a week in 2016, and NACS adopted by GM and Ford. Best direct evidence of pressure: Diess at VW, August 2017. But VW's moves were also Dieselgate ($2bn settlement) and CO2 rules.
⚠️ Two myths the report kills: the patent pledge is conditional good faith, not a licence or public domain, and citations do not prove use; profitability came late, the first fully profitable year being 2020 at $721 million after a $465 million DOE loan in 2010.
So both podcasts already have their research. The second one even has its structure: the report recommends a compact package of 18 primary sources and a 35 to 45 minute episode with the verdict stated plainly, Tesla as neither originator nor passenger but one of the important accelerators of BEV competition.
What is genuinely left is the editorial work: keeping the four safety questions separate, labelling every "N times safer" claim as non-causal, comparing Robotaxi with Waymo only within a similar operational domain and the same denominator, and not letting employee and union testimony stand in for evidence about the cars.
The full story, in two versions:
📖 For humans, the longread: https://github.com/tonydzi/clawrush/blob/main/longreads/20261103.md
🤖 For machines, the devlog: https://github.com/tonydzi/clawrush/blob/main/devlog/20261103.md. Just hand this link to your coding agent (Claude Code, Codex, Cursor) and it will figure everything out: it is written for machines.
🔗 All our channels and contacts in one place: https://linktr.ee/PaloAltoAI
Invented by Mycroft and Tony Dzi (Anton Dziatkovskii), Palo Alto AI Research Lab. Proudly made in Silicon Valley.
Top comments (0)