Tesla FSD 14.3.9 Can Take Control During Manual Driving to Avoid a Crash

Tesla is giving Full Self-Driving a new job even when the person behind the wheel has chosen to drive manually.

FSD Supervised v14.3.9 introduces Automatic Collision Evasion, a safety feature designed to activate if the car detects an imminent frontal crash and determines that braking alone may not be enough.

Automatic Emergency Braking tries to stop the car in its lane. Tesla’s new system may call on the broader FSD stack to steer, brake or accelerate toward a safer path.

The feature is not a new autonomous-driving mode, and broad owner availability has not begun. It moves part of FSD from something a driver deliberately starts into something that can wait in the background during manual driving.

Tesla AI announced the new safety layer and identified three situations behind it: an imminent collision that braking may not solve, heavy driver distraction, or an accidental disengagement of FSD.

FSD Supervised v14.3.9 starting to roll out shortly

This release includes a new active safety feature set: FSD Supervised can now activate on your behalf when an imminent collision is detected and Automatic Emergency Braking (AEB) may not be enough.

It may also engage if we…

— Tesla AI (@Tesla_AI) September 4, 2026

Tesla is targeting the cases where braking cannot clear the vehicle’s path. Some crash threats demand a turn as well as a rapid reduction in speed.

A blocked lane, a rapidly closing obstacle or a driver who has stopped responding can demand a steering decision as well as deceleration.

Not a Tesla App’s 2026.27.5 release notes identify the feature as Automatic Collision Evasion and say it can engage while the vehicle is being driven manually. The listing describes two main trigger groups: an imminent frontal collision that braking alone may not avoid, and a situation where the vehicle sees inadequate driver attention or believes FSD was switched off unintentionally.

The current listing identifies AI4 hardware and clearly includes Model 3 and Model Y. Model S, Model X and Cybertruck appear with uncertainty markers, so owners should not assume support from the headline alone.

Availability can vary by hardware, model and software branch.

The same tracker showed no fleet installations when this article was prepared. Tesla said the release was starting shortly; broad public rollout had not begun.

Owners should wait for the update and the in-car release notes on their own vehicle.

The collision feature leads the release, but v14.3.9 also carries a long list of changes underneath it.

The notes describe a new AI compiler and runtime that reduce reaction time by 20 percent, upgrades to the neural-network vision encoder, and more training around low-visibility scenes, emergency vehicles, school buses, unusual objects, small animals and complicated traffic lights. Tesla also says it reduced minor tailgating and unnecessary lane bias while improving driver-monitoring sensitivity.

Tesla provided the architecture behind those claims in its first-quarter 2026 update. The company said v14.3 upgraded the reinforcement-learning stage for difficult edge cases, sharpened the vision encoder’s low-visibility understanding and rebuilt the compiler and runtime to cut inference latency by as much as 20 percent.

The April update placed those changes inside Tesla’s larger autonomy program, which spans customer vehicles and the Robotaxi fleet. Tesla said v14.3 had launched that month and described the lower-latency runtime as a way to speed both on-road decisions and the company’s own model-development cycle.

Tesla also tied the software work to a new in-vehicle Self-Driving app for AI4 vehicles, where owners can subscribe, learn how to use FSD Supervised and review driving statistics. The company kept the central limitation in the same investor document: active driver supervision remains required, and the software does not make the vehicle autonomous.

That lower latency matters most when time is disappearing. Collision evasion does not get seconds to deliberate.

Detection, path choice and vehicle response all have to happen quickly enough to change the outcome.

The release also shows how Tesla is trying to turn one driving stack into several products. Its notes say the model is unified across FSD, Robotaxi and Actually Smart Summon.

Automatic Collision Evasion extends that same system into another role: an emergency intervention layer behind a human driver.

NHTSA describes conventional Automatic Emergency Braking as a system that detects an impending collision with a vehicle or pedestrian and applies the brakes when the driver has not. The agency’s federal standard is built around reducing forward crashes through braking and is expected to become mandatory on nearly all new U.S. light vehicles by September 2029.

Tesla’s feature goes beyond that baseline by adding directional control when the company believes a stop in the current path will not be enough. That wider authority is exactly what makes the idea potentially valuable—and why the quality of its judgment will matter so much.

A well-timed maneuver could avoid an obstacle that no amount of straight-line braking could miss. A poorly timed or false intervention could surprise the driver at the worst possible moment.

Tesla has not yet published real-world activation rates, false-positive data or broad owner results for this release.

Tesla continues to say that FSD Supervised requires active driver supervision, does not make the vehicle autonomous and does not replace the person behind the wheel. Drivers must remain attentive and ready to take control, even as the software gains more authority to act.

That warning is not boilerplate here. Automatic Collision Evasion is designed for the moment when normal control is already breaking down, whether because a crash is developing, the driver is badly distracted or FSD appears to have been disengaged by mistake.

FSD 14.3.9 therefore represents more than another collection of smoother turns and better parking behavior. Tesla is testing a bigger proposition: that the same vision-based system used for supervised driving can make a better split-second escape decision than brakes alone.

The first release notes tell owners what Tesla intends. The real measure will come after the software reaches cars and begins making those decisions on actual roads.