Tesla is recalling 362,758 U.S. vehicles equipped with or awaiting installation of its Full Self-Driving Beta software after federal safety regulators concluded that the system can create an unreasonable crash risk in certain roadway situations, including intersections, stop signs and changes in posted speed limits. The remedy will be delivered through a free over-the-air software update rather than a physical repair, making the action both a conventional federal safety recall and an unusually visible test of how regulators oversee software-defined vehicles.

The National Highway Traffic Safety Administration assigned the action campaign number 23V-085. Its Feb. 16 recall notice says affected vehicles include certain 2016–2023 Model S and Model X cars, 2017–2023 Model 3 cars and 2020–2023 Model Y vehicles. NHTSA said the software may allow vehicles to proceed straight from a turn-only lane, enter a stop-sign-controlled intersection without fully stopping, move through a yellow signal without sufficient caution or respond inadequately to posted speed-limit changes.

The defect centers on traffic-law compliance

The recall does not allege that Full Self-Driving Beta fails in every use or that the vehicles are autonomous. It focuses on a defined set of behaviors in which the software can act unlawfully or unpredictably. NHTSA says those behaviors increase crash risk, and federal law requires manufacturers to remedy safety-related defects once identified.

Tesla agreed to conduct the recall but does not agree with NHTSA’s analysis, according to contemporaneous Reuters reporting. The company told regulators it was not aware of deaths or injuries associated with the specific defect and had identified 18 warranty claims that could relate to the issue. The planned software update is expected to revise how the system negotiates the identified situations.

A TechCrunch account of the recall says NHTSA raised the concerns with Tesla in late January and the two sides met repeatedly before Tesla filed the defect report. The chronology illustrates a regulatory model increasingly important for connected vehicles: agencies can identify software behavior through testing and investigation, manufacturers can modify code remotely, and the legal recall process still applies even when no vehicle needs to enter a service center.

“Full Self-Driving” remains a driver-assistance system

Despite its name, Full Self-Driving Beta does not make a Tesla autonomous. Drivers are required to remain attentive and ready to take control. The distinction is central to both safety and consumer understanding because Tesla markets capabilities that can steer, change lanes, respond to traffic controls and navigate city streets while still assigning ultimate responsibility to the human driver.

An Associated Press report on the recall emphasized that the system can perform a broad set of driving tasks but continues to require active supervision. The report also places the action within wider NHTSA scrutiny of Tesla’s automated-driving systems, including investigations involving crashes with emergency vehicles.

Tesla’s own annual Form 10-K, filed at the end of January, identifies Autopilot and Full Self-Driving as important software capabilities and also discloses regulatory and legal scrutiny. The filing says the Justice Department has requested documents related to those systems, without stating that any authority has concluded Tesla committed wrongdoing.

That disclosure became public before the recall and was covered by the Associated Press in a separate report. Together, the federal inquiries show that Tesla’s automated-driving strategy is being examined through several channels: defect investigations, crash data, marketing claims and broader law-enforcement requests.

Software updates do not remove the recall framework

Chief Executive Elon Musk has objected publicly to use of the word “recall” for a problem corrected remotely, arguing that the term is outdated when a software update can be installed without a service visit. Federal regulators use the term differently. Under U.S. safety law, a recall describes the legal process for correcting a safety defect, not the physical method used to fix it.

That distinction matters as cars increasingly resemble networked computing platforms. A defect can be introduced, modified or corrected through software after a vehicle has been sold, but the safety consequences remain physical. A faulty decision at an intersection can produce the same collision risk whether the underlying cause is mechanical hardware or code.

NHTSA has also required manufacturers and operators to report certain crashes involving advanced driver-assistance and automated-driving systems under a standing order first issued in 2021. Those data are intended to give regulators earlier visibility into safety patterns as the technology spreads.

The recall tests Tesla’s rapid-deployment model

Tesla has used customer fleets to deploy and iterate Full Self-Driving Beta at a scale unusual in the auto industry. Supporters argue that real-world data allow the software to improve faster than limited closed-course testing. Critics contend that public roads expose ordinary drivers, pedestrians and other road users to software that is explicitly still in beta.

The 362,758-vehicle recall brings that tension into the federal defect process. NHTSA is not banning Full Self-Driving Beta, and the action addresses a specified set of behaviors rather than the entire system. But it establishes that software performance can cross the line from an imperfect beta feature into a legally recognized motor-vehicle safety defect.

As of Saturday, Tesla plans to issue the remedy over the air and notify owners by mid-April. NHTSA’s broader Autopilot investigation remains active. The larger question is no longer whether advanced driver-assistance software will be regulated like other vehicle systems; this recall shows that it will. The unresolved issue is how quickly oversight can keep pace with software that changes far more often than the cars carrying it.