Tesla's Cybercab robotaxis, which began public rides in Austin, were involved in eight accidents in under two months. According to US traffic safety records, the vehicle was in fully autonomous mode in four of the incidents, while a human-supervised driving system was engaged in the other four. This split tests Tesla's claim that "safety remains the same level in unsupervised mode" and reopens the debate over autonomous robotaxis.

What happened?

Shortly after the Cybercab fleet launched commercial service in Austin, it was involved in eight accidents. One person injured in the incidents was taken to the hospital. According to the records, exactly half of the eight incidents occurred while the vehicle's AI-powered driving software was operating without human intervention; the remaining half were recorded in models still under safety supervision.

Why does it matter? How is Tesla's autonomy claim being tested?

For years, Tesla has developed its Full Self-Driving software using a camera-based, lidar-free approach, arguing that this approach is easier and cheaper to scale. The company's marketing language holds that autonomous driving is statistically safer than a human driver. Cybercab's Austin launch was the first major public test of that claim. However, the fact that half the accidents occurred in fully autonomous mode and half in models that still require human supervision shows that even the company's own fleet is segmented by level of autonomy. That split points to an operational reality that contradicts the "safety is the same whether or not there's supervision" messaging.

The comparison with competitors is telling here. Waymo, which uses a multi-sensor approach combining lidar, radar, and cameras together, pursued a gradual, city-by-city expansion, commercializing its operations in San Francisco and Phoenix only after years of testing and data collection. GM's Cruise, on the other hand, lost its operating permit in San Francisco in 2023 after an accident in which a pedestrian was dragged by one of its vehicles — an incident that brought the company's robotaxi business to a near-total halt. Tesla's approach differs from both examples: a narrower sensor set, a faster commercialization timeline, and taking advantage of Texas's relatively loose regulatory framework for approvals.

Who does this affect, and how?

The even split of accidents across autonomy levels is open to several readings. First, the sample is small: drawing a conclusion of "risk doesn't increase with autonomy" or "risk does increase" from eight accidents rests on weak statistical grounds. Second, the fact that accidents also occurred in human-supervised models shows that a safety driver can't prevent every mistake — a reminder that supervision itself isn't an absolute safety net.

CompanySensor approachExpansion strategyStatus after critical incident
Tesla CybercabCamera-based, no lidarFast, single-region commercializationReview at an early stage
WaymoCamera + lidar + radarGradual, city by cityNo large-scale suspension experienced
Cruise (GM)Camera + lidarHalted after rapid expansionPermit revoked after 2023 accident, operations stopped

The impact on the consumer side is direct: a rider in Austin most likely has no idea whether the vehicle is operating under supervision or in full autonomous mode at any given moment. On the regulatory side, NHTSA's past reviews of Tesla's driving software are well known; the new accident data will likely be factored in as additional input to those files. In terms of insurance and liability, accidents in fully autonomous mode carry the potential to shift fault from the driver to the manufacturer — which means both legal and reputational risk for Tesla.

What's next?

There are three things worth watching: whether NHTSA folds these eight incidents into its existing reviews or opens a new formal investigation, whether Tesla adjusts its expansion timeline beyond Austin in the wake of the accidents, and how transparently the company shares its accident data going forward. As the number of cases grows, the accident rate by autonomy level can be calculated more meaningfully; for now, eight incidents aren't enough to draw a strong statistical conclusion.

In the end, this dataset neither confirms nor refutes Tesla's claim that "the level of autonomy doesn't change safety." What it does show is that the company accelerated its public launch before its oversight and data-collection infrastructure was fully in place.