NHTSA’s Standing General Order crash-reporting page now lists downloadable incident files for automated driving systems (ADS) and Level 2 advanced driver assistance systems (ADAS) covering June 16, 2025 through May 15, 2026. The order requires identified manufacturers and operators to report certain crashes involving ADS- or SAE Level 2 ADAS-equipped vehicles. NHTSA warns the public files are not exposure-normalized, may contain multiple reports for the same crash, and may carry incomplete or unverified information when first submitted. That makes them a governance dataset, not a rate filing: they confirm that serious ADAS-involved crashes are still being reported, but they do not tell a risk manager how many crashes occurred per mile, per unit, or per route type.
For a self-insured fleet, the reserve question is narrower and sharper. It is not whether the federal files prove ADAS works. It is whether the fleet has already cut its bodily injury severity pick or expected loss ratio for driver-assist technology before its own paid and incurred experience supports the credit, and whether the credit it took is the right size. The honest answer is that the most credible third-party evidence supports a frequency credit, not a severity one, and the two are moving in opposite directions.
Who it affects
The first audience is self-insured trucking, delivery, utility, service, and transit fleets that have bought cameras, automatic emergency braking (AEB), lane support, adaptive cruise control, or integrated telematics. Captives writing fleet auto liability are in the same position, especially where the board approved a lower expected claim ratio on the strength of vendor safety projections. Public-entity fleets and hospital systems running ambulances, vans, and service vehicles should make the same distinction between collision avoidance and claim severity.
Policy direction reinforces the long-term safety case but says nothing about a current-year reserve. NHTSA’s April 2024 Federal Motor Vehicle Safety Standard No. 127 makes AEB standard on cars and light trucks by September 2029. A policy that mandates the technology for future model years does not retroactively prove that a specific fleet’s open accident years deserve a severity discount today.
The reserve mechanism, sized
The lever is the severity assumption that feeds commercial-auto loss picks and IBNR, and the trap is that ADAS pushes frequency and severity in opposite directions. The best fleet-specific evidence comes from the IIHS large-truck study, which examined crashes per vehicle mile traveled across 62 carriers running tractor-trailers and other trucks weighing at least 33,000 pounds, over more than two billion miles in 2017 through 2019. Forward collision warning cut rear-end crash rates 44 percent and AEB cut them 41 percent. Those are large, real numbers, and they are a frequency result: across all crash types, the same trucks saw only 22 percent (forward collision warning) and 12 percent (AEB) fewer crashes, because the systems target the specific rear-end collisions they are built to prevent and leave the rest of the loss distribution largely intact.
That distinction is the whole reserve argument. Rear-end frequency is exactly the high-frequency, often lower-severity slice of the book. The crashes that drive a commercial-auto bodily injury tail, intersection, lane-departure, and multi-vehicle events with attorney involvement, nuclear-verdict venues, and late case-reserve development, are not the crashes AEB reliably stops. Industry severity is still climbing through that tail: AM Best’s analysis underlying the line’s reported reserve deficiency puts average commercial-auto liability claim severity at roughly 8 percent annual growth, more than double general inflation and more than doubled over nine years. A 41 percent cut in rear-end frequency does not offset an 8 percent compounding climb in severity on the claims that survive.
There is a second channel that runs the other way on physical damage, and it is now measurable. ADAS-equipped vehicles cost more to repair because sensors, cameras, and radar must be replaced and recalibrated. Per CCC’s Crash Course 2026, calibrations appeared on 28.3 percent of repairable appraisals in 2025, up from 21.8 percent a year earlier, at an average fee of $485.56 per calibration and roughly $350 to $500 added per system. The same data shows newer, ADAS-laden vehicles averaging $5,721 in repair cost against $3,682 for older units. So the technology that may trim some collision frequency simultaneously raises the cost of every claim that does occur. actuary.info has framed this directly as a frequency-severity paradox, noting that ADAS prevents the frequent low-cost collisions while inflating the cost of each claim that still happens, a structural reason additive trend models understate the compounding effect on severity.
For reserve work, the implication is that a severity credit should not start life as a blanket frequency reduction copied onto the loss pick. It should show up first in diagnostics: faster liability decisions, lower attorney-represented bodily injury severity after controlling for territory and vehicle class, shorter open duration, and fewer late case-reserve increases. If those measures have not moved, the expected loss ratio should not move simply because the fleet has newer trucks. And bodily injury and physical damage need separate triangles, because the calibration-driven repair severity documented in our ADAS calibration and parts-inflation coverage can mask, or be masked by, the liability picture if the two are blended, as the commercial-auto fleet IBNR playbook explains.
The call
Most fleets that have taken an ADAS severity credit have taken the wrong one. The defensible position from the data is a measured frequency credit on the specific exposures AEB and forward collision warning actually address, principally rear-end events, and no automatic severity credit at all until the fleet’s own attorney-represented bodily injury experience demonstrates it. The frequency-down, severity-up dynamic means the line’s loss-ratio math can deteriorate even as crash counts fall, which is precisely the pattern showing up in fleets where premium has diverged from crash-rate improvement, as the ATRI fleet-premium-versus-crash-rate divergence illustrates. Telematics and dashcam programs are worth real underwriting recognition, but the Louisiana dashcam-discount debate is a reminder that camera footage cuts both ways: it can close a defensible claim faster, and it can hand plaintiff counsel a contemporaneous record of speed, distraction, or delayed response.
What to watch over the next four quarters: whether NHTSA’s SGO files and IIHS effectiveness work begin to separate severity outcomes from frequency outcomes for heavy commercial units; whether calibration’s share of repair appraisals keeps climbing past 28 percent and pulls physical-damage severity up with it; and whether your own ADAS-equipped cohort shows a genuine, controlled reduction in bodily injury incurred severity rather than only fewer fender-benders. The NHTSA files are useful precisely because they are imperfect. They remind reserve committees that public crash data, vendor safety claims, and actuarial selections answer different questions, and that a severity credit belongs in the reserve study only when the fleet’s own claim experience earns it.
Sources
- NHTSA Standing General Order on Crash Reporting
- IIHS: Study Shows Front Crash Prevention Works for Large Trucks Too
- NHTSA Finalizes FMVSS No. 127 Automatic Emergency Braking Rule
- CCC Intelligent Solutions: Crash Course 2026
- actuary.info: ADAS Creates a Frequency-Severity Paradox for Auto Insurers
- NHTSA Fatality and Injury Reporting System Tool