Tesla’s latest Full Self-Driving update is pushing driver assistance into territory that goes beyond simply braking when trouble appears. FSD (Supervised) v14.3.9 introduces Automatic Collision Evasion, a feature designed to activate the company’s driving system during certain emergencies even when a human is manually controlling the car. If a frontal crash appears imminent and braking alone may not be enough, the vehicle can attempt an evasive response.
The development is significant, but the wording matters. Tesla still classifies FSD as supervised driver assistance, not autonomous driving, and early appearances of 14.3.9 have involved internal or limited vehicles rather than a confirmed fleet-wide public deployment. The update also arrives while federal regulators continue examining Tesla’s automated-driving performance in difficult roadway conditions.
Automatic Collision Evasion Is the Headline Change
FSD v14.3.9 appears with Tesla software build 2026.27.5, and its biggest addition is called Automatic Collision Evasion. Release information surfaced from vehicles carrying the new software and from trackers documenting Tesla’s internal rollout. The system is unusual because it can activate while the person behind the wheel is driving manually rather than already using FSD. In other words, the driving software may move from being something deliberately switched on to something capable of entering an emergency situation automatically.
That distinction could make 14.3.9 one of Tesla’s more consequential safety updates in recent years. Conventional collision-avoidance technology often concentrates on warnings, reducing acceleration or applying emergency braking. Tesla’s new approach can call upon the broader FSD driving stack when the vehicle determines that simply stopping may not solve the problem. Early rollout data, however, indicate that 2026.27.5 should not yet be described as universally available to Tesla owners.
The Car Can Intervene When Braking May Not Be Enough
Tesla describes one trigger in straightforward terms: a frontal collision is imminent and braking alone may not avoid it. Under those circumstances, Automatic Collision Evasion can activate FSD in an effort to keep the vehicle safe and continue driving. That creates the possibility of steering around a hazard rather than remaining on a collision course while relying exclusively on the brakes. Reporting on the test build indicates the system can combine steering, braking and acceleration during an evasive manoeuvre when its operating conditions are met.
There is also a second potential trigger. Tesla says the feature may engage if the vehicle determines that the driver is not sufficiently attentive or believes FSD may have been unintentionally disengaged. Tesla gives reaching toward the back seat as one example of inadequate attention. That means the technology is not designed solely around detecting an obstacle; it also incorporates information about what the person inside the vehicle is doing at a potentially critical moment.
FSD Still Does Not Turn a Tesla Into an Autonomous Car
The name “Full Self-Driving” can make any expansion of control sound more autonomous than Tesla itself says the system is. Tesla’s current support documentation repeatedly describes FSD as supervised driver assistance. The company says an attentive driver remains responsible for the vehicle, must monitor its operation and must be prepared to take control. FSD can navigate streets, change lanes, negotiate intersections and park, but Tesla explicitly says those capabilities do not make the vehicle fully autonomous.
That distinction remains important with Automatic Collision Evasion. Emergency intervention does not transfer permanent responsibility from the driver to the software. Instead, the new feature effectively adds another layer of active safety that can temporarily employ capabilities normally associated with FSD. Tesla also notes that feature availability can differ according to vehicle hardware, software version, region, model and regulatory approval. A dramatic emergency capability appearing in release notes therefore should not be interpreted as permission for drivers to pay less attention.
This Goes Beyond Tesla’s Existing Emergency Braking
Tesla vehicles already have a group of active-safety functions including Forward Collision Warning, Automatic Emergency Braking and Obstacle-Aware Acceleration. Forward Collision Warning alerts the driver when the vehicle determines there is a high risk of a frontal collision. Automatic Emergency Braking can then apply the brakes if a collision is considered imminent. Tesla’s owner documentation makes an important qualification: emergency braking is intended to reduce collision impact and should not be relied upon to prevent every crash.
Automatic Collision Evasion addresses a different problem. There are circumstances in which slowing rapidly may still leave a vehicle directly in the path of an obstacle. Version 14.3.9 is designed to give the car another option by bringing FSD’s path-planning and steering capability into the response. That is conceptually closer to an automated evasive manoeuvre than traditional emergency braking. Whether that translates into measurable real-world crash reductions will depend on deployment scale, intervention accuracy and how reliably the system distinguishes genuine emergencies from situations where unexpected steering could create another hazard.
Version 14.3.9 Also Changes the FSD Neural Network
Collision evasion is getting most of the attention, but the accompanying FSD update contains a much broader collection of changes. Tesla’s release notes say the reinforcement-learning stage used to train the FSD neural network has been upgraded, targeting improvements across a range of driving situations. The vision encoder has also been revised to improve understanding of rare situations, poor visibility, three-dimensional road geometry and traffic signs.
Another technical change involves the software underneath the neural network. Tesla says it rewrote its AI compiler and runtime using MLIR and claims the change produces a 20% faster reaction time while also allowing faster model iteration. Reaction time can matter enormously when software is deciding whether another car, pedestrian or road object requires immediate action. Still, that 20% figure is Tesla’s software-performance claim rather than an independently established reduction in crashes. Real-world safety performance remains the more important measure as 14.3.9 reaches additional vehicles.
Rare Road Situations Are Getting More Training Attention
Many difficult automated-driving problems are not the ordinary miles where traffic behaves predictably. They are the delivery truck blocking half a lane, the school bus stopping unexpectedly, an emergency vehicle approaching from an unusual direction or an object hanging into the roadway. Tesla says v14.3.9 improves responses to emergency vehicles, school buses, vehicles violating right of way and other relatively uncommon road users or situations.
The update also targets smaller and less predictable hazards. Tesla says reinforcement-learning work has been focused on handling small animals more safely and on unusual objects extending, hanging or leaning into a vehicle’s path. Traffic-light behaviour at complicated intersections is another area receiving additional training, including compound signals, curved roads and decisions around yellow lights. These additions illustrate why automated driving is difficult to perfect: ordinary driving can be learned from enormous amounts of repetitive data, while low-frequency events may require deliberately identifying and emphasizing the rare examples that can become disproportionately dangerous.
Low-Visibility Improvements Arrive Under Regulatory Scrutiny
The release notes say v14.3.9 improves the system’s understanding of rare and low-visibility situations. That is particularly notable because reduced visibility is already the subject of a major National Highway Traffic Safety Administration investigation. In March 2026, NHTSA upgraded its review to an Engineering Analysis covering an estimated 3.2 million FSD-equipped Teslas after examining crashes involving glare and other degraded-visibility conditions.
NHTSA said its investigation was examining whether Tesla’s degradation-detection system could recognize impaired visibility and warn drivers with enough time to react. The agency’s case summary listed nine crashes, including one fatal incident, within the investigation’s failure-report data. An open investigation does not establish that every vehicle or FSD version contains a defect, and v14.3.9 is newer than much of the software being examined. Nevertheless, the overlap is significant. Tesla is specifically emphasizing low-visibility understanding at a time when regulators are scrutinizing how effectively its camera-based system recognizes exactly those conditions.
Driver Monitoring Is Becoming More Important, Not Less
Automatic Collision Evasion’s ability to react to severe inattention puts additional importance on the cabin-monitoring system. Tesla says v14.3.9 improves driver-monitoring sensitivity through better eye-gaze tracking, improved handling of eyewear and greater accuracy under changing lighting conditions. The system therefore has two complicated tasks occurring simultaneously: interpreting what is happening outside the car and determining whether the person inside is sufficiently engaged.
That matters because FSD remains a supervised system. A driver-assistance technology can become less safe if motorists misunderstand its capabilities or stop watching the road because the software usually performs well. Tesla continues to warn against complacency and says drivers remain responsible for speed and control even when FSD is operating. Version 14.3.9 does not eliminate that human-machine relationship. Instead, the emergency-intervention feature arguably makes it more complex, because the vehicle may now intervene during manual driving precisely when its monitoring systems conclude that immediate human action may not arrive.
Tesla Reports Strong Safety Numbers, but Context Still Matters
Tesla says vehicles operating with FSD Supervised experience fewer collisions than Teslas driven without it. Its current Vehicle Safety Report displays more than 11 billion cumulative FSD miles and says FSD operation is associated with seven times fewer major collisions, seven times fewer minor collisions and five times fewer off-highway collisions under the company’s methodology. Tesla uses vehicle telemetry to classify incidents and counts FSD as involved if it was active within five seconds before a collision.
Those numbers provide valuable large-fleet data, but they are produced and analyzed by Tesla rather than an independent regulator. Driving environment also matters when comparing crash rates because supervised automation may be used disproportionately on certain roads, in certain vehicles and by certain motorists. Regulators therefore examine specific failure modes as well as aggregate statistics. NHTSA, for example, separately opened an investigation involving alleged traffic-law violations with FSD engaged, including red-light and wrong-way incidents. The two sets of evidence are not mutually exclusive: overall improvement can coexist with particular behaviours that still need investigation.
The Biggest Test Comes When Ordinary Owners Receive It
For now, the most important limitation surrounding v14.3.9 is rollout status. Reporting on software 2026.27.5 initially described it as an internal Tesla employee build, while software trackers showed little or no broad customer deployment as the version emerged. A vehicle running the software has since been spotted outside the most obvious development setting, but that still does not establish a normal fleet-wide release. Tesla frequently stages updates gradually before expanding availability.
That cautious rollout will matter more than usual with Automatic Collision Evasion. A parking improvement that behaves awkwardly is inconvenient; emergency steering that intervenes incorrectly carries a different level of consequence. Engineers must balance the danger of failing to act against the danger of acting when intervention was unnecessary. If 14.3.9 progresses successfully through early testing, Tesla will have effectively expanded FSD from a driver-selected assistance system into an active-safety layer capable of stepping into certain emergencies. That is a substantial change even though the driver officially remains responsible.
