A fatal intersection crash in southern New Jersey is drawing fresh scrutiny more than a year later because Tesla’s federal crash filing adds a detail that was absent from the initial public account: a driver-assistance system was listed as “Verified Engaged.” The July 2025 collision in Buena Vista Township killed 82-year-old Stephen Field after a Tesla Model 3 failed to stop at a stop sign and struck his Honda Civic.
Yet the public federal record does not identify the exact Tesla software version or provide the full crash narrative. That gap matters. It means the available evidence can establish that automation was active during the critical window, but not whether the software caused the stop-sign violation. The result is a case that sits at the center of a larger debate over driver responsibility, software transparency and how much the public can learn from federally reported automated-driving crashes.
A Routine Crash Report Gained a Second Life
When New Jersey State Police responded to County Route 671 and Chestnut Avenue in Buena Vista Township at about 6:57 p.m. on July 6, 2025, the basic facts appeared straightforward. Police said a Tesla Model 3 failed to stop at a stop sign and struck a Honda Civic that was making a left turn. Stephen Field, 82, suffered fatal injuries. The Tesla’s driver and three passengers were reported to have moderate injuries and were taken to a hospital.
What changed the significance of the case was not a new eyewitness account, but a federal crash-data entry submitted by Tesla. Reporting based on NHTSA’s Level 2 crash dataset identified a 2019 Model 3 in Buena Vista Township and showed its automation engagement status as “Verified Engaged.” That does not overturn the police account. It adds a technical layer that was missing from the original public description and raises an important question: what exactly was the driver-assistance system doing before impact?
“Verified Engaged” Is Meaningful, but Limited
NHTSA’s data dictionary gives “Verified Engaged” a specific meaning within its crash-reporting system. It is one of the engagement-status choices used by a reporting entity to describe the highest level of driving automation active during the 30 seconds before a crash begins through its conclusion. The alternatives include “Alleged Engaged,” “Verified Not Engaged” and “Unknown.” Tesla’s reported designation therefore carries more weight than an unconfirmed statement that a driver believed an assistance feature was operating.
Still, the label is not a federal determination that automation caused the collision. NHTSA warns that reports submitted under its Standing General Order can initially contain incomplete information and material that the agency has not independently verified. They are designed to alert regulators to incidents that may warrant further examination, not serve as completed crash reconstructions. In this case, engagement status narrows the technical questions, but responsibility still depends on evidence such as telemetry, driver inputs, roadway conditions, braking and the complete sequence of events.
The Most Important Software Details Are Still Hidden
The publicly available record reportedly withholds the automation feature version, the crash narrative and the field indicating whether the vehicle was operating within the system’s intended operational design domain. NHTSA permits reporting entities to seek confidential treatment for those three categories. When a confidentiality request is properly submitted, the agency can withhold the information from its public dataset while the claim is handled under federal disclosure rules.
That creates a striking contrast in this crash: the public can learn that a driving-assistance system was active without seeing some of the information that could explain what that system was capable of doing. Version information can help establish which generation of Tesla software and which system behavior applied at a particular time. The narrative could provide additional context unavailable from individual database fields. Without those details, outside observers cannot reliably determine from the public record alone whether Full Self-Driving, a more limited assistance configuration or another combination of features governed the vehicle’s behavior.
Autopilot and FSD Are Not Interchangeable
Tesla’s own documentation separates its assistance features by capability. Full Self-Driving (Supervised) is described as capable of following routes, negotiating intersections, making turns and responding to stop signs and other traffic controls. Tesla describes Autosteer and related functions more narrowly, while some Model 3 configurations have also offered Traffic Light and Stop Sign Control. Hardware, software version, vehicle configuration and purchased options can all affect which functions are available on a particular car.
That distinction is critical here. A stop-sign crash might naturally prompt speculation about FSD because that system is designed to operate through city streets and intersections, but the publicly released evidence does not establish that FSD was the active feature. Nor would the presence of another driver-assistance function, by itself, reveal exactly how the vehicle approached the intersection. Tesla repeatedly warns that its assistance technology is not autonomous and that drivers must be ready to intervene. Until the feature and software version are established, broader labels such as “Autopilot” risk obscuring more than they explain.
The Reported 4 MPH Figure Invites Comparison — Not a Conclusion
Reporting on the federal entry says Tesla listed a pre-crash speed of 4 mph. The figure attracts attention because Tesla previously recalled 53,822 vehicles over a Full Self-Driving Beta feature capable of making a so-called rolling stop. In the 2022 recall, NHTSA said the functionality could allow affected vehicles to travel through an all-way-stop intersection without first coming to a complete halt. Tesla’s defect filing described strict operating conditions and a maximum rolling-stop speed of approximately 5.6 mph.
The comparison is noteworthy, but it does not establish that the earlier defect played any role in the Buena Vista collision. Tesla disabled the rolling-stop functionality through an over-the-air software update years before the 2025 crash. The software version associated with the New Jersey vehicle also remains hidden from the public record. More importantly, one speed field cannot reconstruct a collision. Investigators would need to know precisely where that speed was measured, how braking changed, whether the accelerator was used, what the driver did and when the Tesla entered the intersection.
Level 2 Automation Still Requires an Attentive Driver
The federal framework treats systems such as Tesla’s FSD (Supervised) as Level 2 driver assistance rather than fully autonomous driving. At Level 2, technology can simultaneously provide steering and acceleration or braking assistance, but the human driver remains responsible for supervising the vehicle and must be ready to intervene. Tesla makes the same point in its own documentation, stating that its self-driving features do not turn a Model 3 into an autonomous vehicle or replace the person behind the wheel.
That structure becomes particularly important after a serious collision. Investigators can examine whether the software issued appropriate commands and warnings while separately asking whether the driver monitored the road and had an opportunity to respond. Those inquiries can overlap rather than cancel each other out. A driver may retain legal and safety responsibilities even if software performs unexpectedly, while a manufacturer’s system can still face scrutiny if its behavior contributed to danger. In Buena Vista, the public evidence does not yet establish how those responsibilities should be divided.
Federal Regulators Are Already Examining Traffic-Law Violations
The New Jersey collision comes against the backdrop of a separate, open NHTSA investigation into Tesla’s Full Self-Driving technology and alleged traffic-law violations. The agency opened Preliminary Evaluation PE25012 in October 2025 to examine whether FSD-equipped vehicles can perform maneuvers that violate traffic signals, signs or lane markings, and whether drivers receive adequate warning and sufficient opportunity to react. The inquiry includes allegations involving red traffic signals, travel into opposing lanes and other improper roadway movements.
When the investigation opened, NHTSA said it had identified 18 complaints and one media report concerning alleged failures around red traffic signals. The agency also identified six crash reports in which Teslas operating with FSD entered intersections against red lights, with four of those collisions involving injuries. None of that proves a defect, and the Buena Vista stop-sign crash should not automatically be treated as part of that investigation. The broader relevance is narrower: regulators are already examining how Tesla’s supervised Level 2 technology behaves around basic traffic-control requirements.
NHTSA’s Crash Database Is Powerful, but Not a Scoreboard
NHTSA’s Standing General Order was designed to give federal safety officials timely information about crashes involving automated-driving systems and Level 2 advanced driver assistance. For Level 2 vehicles, qualifying crashes generally must involve use of the assistance technology within 30 seconds of the event and meet reporting criteria such as a fatality, hospital treatment, air-bag deployment or involvement of a vulnerable road user. That reporting system can expose technical details that never appear in routine local crash coverage.
But the agency explicitly warns against treating the database as a simple ranking of manufacturers or automated-driving technologies. Companies have different access to telemetry, connected-vehicle notifications and customer reports. One manufacturer may automatically learn about many crashes while another may depend heavily on people notifying it. The figures also are not normalized by fleet size or miles driven. As a result, raw crash totals cannot establish which system is safer. In the Buena Vista case, the database is most valuable as a trail toward evidence, not a statistical verdict.
The Redactions Are Part of a Larger Transparency Fight
The missing Tesla information is not merely a quirk of database formatting. Tesla has sought confidential treatment for categories including automation-version information and crash narratives, arguing that public disclosure could reveal proprietary information and harm its competitive position. That issue reached federal court in litigation involving The Washington Post and NHTSA. In March 2026, a judge found that the government had adequately supported withholding Tesla’s version and narrative information under the Freedom of Information Act’s Exemption 4, which protects qualifying confidential commercial information.
The court was less persuaded by the government’s justification for withholding Tesla’s operational-design-domain information, illustrating how contested these transparency questions have become. For families, researchers and safety advocates, the practical consequence is significant. Federal regulators can possess far more detail than the public sees. A crash record may disclose that automation was active while concealing the exact version, operating conditions and explanatory narrative. With vehicle behavior increasingly determined by frequently changing software, those omitted details can separate informed analysis from speculation.
The Remaining Questions Require Vehicle Data, Not Guesswork
The most useful next step is not an assumption that Autopilot or FSD must have caused the crash. It is a detailed reconstruction using vehicle data, driver inputs, physical evidence and the police investigation. Reporting on Tesla’s federal filing says event-data-recorder information and telematics were available. Depending on the information preserved, those records could help establish how vehicle speed changed, when assistance was engaged, whether braking or accelerator inputs occurred and whether the driver intervened before the collision.
Until investigators release findings or more underlying data becomes public, several explanations remain possible. The assistance system could have behaved improperly around the stop sign. The driver could have overridden, misused or failed to supervise it. Human behavior and software performance could both have contributed, or the fact that assistance was engaged could ultimately prove less significant than it initially appears. The verified disclosure is therefore important but narrow: Tesla reported that driver assistance was active during the relevant pre-crash window. What role that technology actually played in Stephen Field’s death remains unresolved.
