Overview
Tesla Autopilot is a driver-assistance system. Tesla Full Self-Driving (FSD) is a more advanced driver-assistance system. Neither one drives itself, though Tesla is now trying to prove otherwise in public. The names have been lies for a decade, and people have died because of those lies.
Elon Musk first promised fully autonomous Tesla vehicles by 2017. Then 2018. Then 2020. Then “next year” every year after that. Customers paid up to $15,000 for FSD capability, and what they received was a beta feature that requires constant driver supervision, can’t handle construction zones reliably, and has been linked to hundreds of crashes and dozens of fatalities.
In January 2026, Tesla started removing the human safety monitor from some Austin robotaxis, so a subset of paid rides now operate with nobody inside able to grab the wheel. NHTSA has investigated Tesla’s driver-assistance systems in connection with over 900 crashes, including incidents where Teslas drove into parked emergency vehicles with flashing lights. In 2023, Tesla recalled 2 million vehicles to update Autopilot’s driver monitoring system, the largest recall in the company’s history. In 2024, another recall of 2.2 million vehicles addressed insufficient warnings.
The fundamental issue isn’t that the technology is imperfect. All technology is imperfect. The issue is that Tesla markets an imperfect system with a name that implies perfection, to customers who lack the training to understand the difference, and is now selling rides in it to strangers who never signed up for the beta at all.
What It Knows About You
Every Tesla is a rolling surveillance platform. The car has eight cameras recording continuously. Tesla collects video footage, GPS data, speed data, braking data, battery data, and cabin camera footage that monitors whether you’re paying attention to the road.
Tesla’s privacy policy allows the company to share data with “service providers and business partners.” In 2023, Reuters reported that Tesla employees privately shared sensitive images captured by car cameras, including images from inside customers’ garages and private property. The images were shared in internal group chats for entertainment.
The robotaxi fleet adds a layer on top of that. Tesla has built a teleoperation system, reportedly including a VR rig, so a remote employee can see a stuck car’s surroundings and steer it clear over a cellular link. Tesla’s robotaxi privacy notice says the cabin camera and microphone stay off during a ride unless a passenger calls for support, and that this footage isn’t tied to the rider’s account. That’s a genuine carve-out, and it’s also a promise that rests entirely on Tesla’s own word, with no independent audit of when the camera actually activates.
If you own a Tesla, the company knows everywhere you drive, how fast you drive, how aggressively you brake, where you park at night, and (via cabin cameras) who’s in the car with you and what they’re doing. That’s not a car. That’s a surveillance vehicle you make payments on.
The Real Risks
Safety is the existential risk here, and the regulatory picture got worse this year. NHTSA escalated its probe into FSD’s handling of sun glare and fog to an Engineering Analysis in March 2026, the stage that typically precedes a recall, covering roughly 3.2 million vehicles. The agency tied nine crashes, including one fatality, to the visibility flaw, and said Tesla’s own data gaps “could have led to under-reporting” of how often it happened. Separately, a Florida jury’s $243 million verdict against Tesla over a fatal 2019 Autopilot crash survived the company’s bid to overturn it in February 2026. Tesla has since settled at least four more wrongful-death Autopilot and FSD suits, including a fatal FSD pedestrian crash in June and a fire truck collision in September, rather than let another jury see the evidence.
The autonomy score moved to 10/10 because Tesla removed the safety net, not just the marketing hedge. Starting in January 2026, some Austin robotaxis began operating without a human safety monitor on board at all. When FSD is engaged in a personal car, the driver still has seconds to intervene if it makes a wrong decision. In a driverless robotaxi, nobody in the vehicle can intervene; the only backup is a remote teleoperator watching a video feed, and every disclosed teleoperator takeover so far, in a July 2025 curb strike and a January 2026 barricade collision, ended in a crash. The automation paradox that made this risky in personal cars, that the better a system performs on average, the worse humans respond in the rare moments it fails, now runs with no human in the loop able to respond at all.
Job threat rose from 7 to 8 once the robotaxi moved from pilot to paid product. Personal FSD assists a driver who is still employed doing something else. A driverless robotaxi is a taxi, competing directly with professional rideshare and cab drivers for the same fares, and Tesla has said expanding this fleet, including a bid for thousands of robotaxi slots in Las Vegas, is central to the company’s plans. That is a live displacement, not a roadmap slide.
Lock-in is brutal. FSD is tied to the vehicle, not the owner. If you sell your Tesla, you lose the $15,000 feature. If Tesla decides to change the terms (which they have), you have no recourse. The car’s software is controlled entirely by Tesla via over-the-air updates. They can add features, remove features, or change how existing features work without your consent. In 2023, Tesla remotely reduced the range of some older vehicles via software update.
Bias matters in autonomous vehicles. Studies have shown that computer vision systems perform worse on darker skin tones, meaning pedestrian detection may be less reliable for people of color. Tesla uses a vision-only approach (no LIDAR, no radar in newer models), making it more susceptible to these biases.
Alternatives
- Don’t use FSD. Use basic Autopilot (adaptive cruise control plus lane keeping) on highways with full attention. It’s far less risky.
- Comma.ai / openpilot: Open-source driver assistance. You can audit the code. It’s honest about its limitations.
- Waymo: Has run driverless robotaxis for longer than Tesla and discloses crash data more consistently, without an NHTSA probe escalating to an engineering analysis. Not risk-free, but a longer track record.
- Public transit: Zero autonomy risk. Zero privacy risk. Gets you there without anyone dying because an algorithm confused a white truck for the sky.
- Other automakers’ ADAS: GM’s Super Cruise, Mercedes’ Drive Pilot, and BMW’s systems are arguably more conservative and honest about their capabilities. Mercedes actually accepts legal liability when Drive Pilot is engaged, Tesla does not.
Our Verdict
Tesla Autopilot / FSD stays an F, and this refresh found the case got worse, not better. NHTSA’s visibility probe is now one step from a recall, with regulators saying Tesla’s own data problems likely hid how often the flaw caused crashes. A $243 million jury verdict over a 2019 death survived Tesla’s appeal, and the company has since paid to settle at least four more wrongful-death suits rather than let a jury see the evidence again. And in January 2026, Tesla started removing the one human who could grab the wheel, sending some Austin robotaxi passengers down public roads with nobody aboard able to intervene if the car gets it wrong; the teleoperators who back them up have crashed every time they’ve had to take over. Autonomy moves to 10/10 and job threat to 8/10 to reflect that. What would move the grade: a completed NHTSA recall that Tesla actually fixes rather than settles around, a halt on removing safety monitors until the visibility flaw is resolved, and prompt, complete crash reporting instead of the delayed batches regulators are now investigating.