Tesla’s FSD Streaks Are Gamifying FSD and Encouraging Risky Behavior [Opinion]

Tesla’s Full Self-Driving (FSD) improves through a feedback loop that combines human supervision with machine learning. The system uses end-to-end neural networks trained on large volumes of real-world telemetry and simulated data, and today human drivers play two roles.

First, they act as safety supervisors who remain legally responsible for the vehicle’s actions. Second, when they intervene manually, those actions serve as annotations that flag situations for further training.

With FSD v14.2, and later FSD v14.3.3 introduced alongside the 2026 Spring Update, Tesla adjusted how drivers interact with this loop. By introducing incentives to avoid disengagements through Distance Streak and Daily Streak in the Self-Driving App, the design encourages specific driver behaviors.

This shift changes how FSD is perceived and may affect both on-road safety and the data processes that have helped make the system safer.

High Score Illusion

The transition began quietly with FSD v14.2, which added a passive metric showing the exact percentage of total miles driven under autonomous control. Although tucked away in a menu, it prompted some owners to aim for a 100% autonomous miles-driven reading.

The 2026 Spring Update amplified this by showing the owner’s maximum distance between interventions directly beneath the active speed indicator, effectively turning a safety-related measure into a running high score.

Any manual override—pressing the brake, tugging the steering wheel, or pressing a button—resets that distance counter to zero, creating a perceived penalty for intervening.

Meanwhile, using the accelerator pedal or turn signals to guide FSD does not reset the counter, drawing a narrow line around what counts as an “acceptable” input. There are no explicit warnings or disclaimers for this feature, nor an option to disable it.

Gambling on “Wait and See”

Although FSD has reached statistical safety levels that exceed the average human driver, it can still encounter edge cases. Even Unsupervised Robotaxis in Austin rely on remote helpers to address issues as they arise.

Recurring problems reported by some owners include erratic path planning, misjudged lane changes, incorrect highway exits, and excessive hesitation or brake-stabbing at stop signs and shadows.

Safety Reality Check

A navigation mistake can still escalate into a hazardous event. For example, taking a runaway truck ramp at highway speeds on FSD was reported in a Cybertruck running FSD v14.3.3, the latest FSD build.

Under typical SAE Level 2 supervision, a driver who notices an error is expected to take control promptly, which helps maintain traffic flow and reduce risk to everyone nearby. The Distance Streak mechanic alters those incentives.

When a multi-hundred-mile streak is on the line, some drivers may delay taking over in the hope that FSD recovers on its own. In a video, the Model Y that embarked on a coast-to-coast FSD trip was nearly totaled.

Giving FSD Bad Data

Beyond immediate safety concerns, gamification can reduce the quality of training data. Tesla’s data flywheel processes millions of hours of driving, including interventions, edge cases, voice notes, and other statistics, to refine its neural networks.

Manual disengagements communicate that FSD did something illegal, uncomfortable, incorrect, or dangerous, and those moments help map the system’s boundaries and highlight what needs improvement to reach Unsupervised FSD. If drivers avoid disengaging in uncomfortable or suboptimal situations, fewer of these valuable signals are collected.

By publicly resetting a driver’s streak after any manual disengagement, the system discourages the very behavior its AI teams depend on. When drivers suppress natural safety interventions to protect a numerical streak, they provide less high-value data to the training pipeline.