Since mid-2021, NHTSA's Standing General Order has required automakers and tech companies to report crashes involving Level 2 driver-assist systems and higher-level automated driving systems (ADS) within 1 to 5 business days, depending on severity. The resulting public dataset has logged tens of thousands of Level 2 reports and a much smaller number of ADS (robotaxi-type) reports. The data is useful for spotting patterns but is not a clean safety scorecard — it has no denominator of total miles driven per system, so it cannot show a crash *rate*, only a raw count of reported incidents.
At a glance
| Detail | What it means |
|---|---|
| What it's called | NHTSA Standing General Order (SGO) on Crash Reporting, first issued 2021, amended since |
| Who must report | Manufacturers and operators of Level 2 ADAS and Level 3-5 ADS involved in a qualifying crash |
| Reporting window | As fast as 1 calendar day for the most severe crashes, up to monthly summaries for minor ones |
| Public access | Searchable dataset published on NHTSA.gov, updated periodically |
| Biggest limitation | No exposure-mile denominator, so raw counts can't be turned into a fair crash rate |
Why this database exists
Before 2021, there was no standardized, mandatory way to know how often ADAS or automated driving systems were involved in a crash — reporting was voluntary, inconsistent, and scattered across individual company disclosures. NHTSA's Standing General Order changed that by compelling manufacturers and companies testing or deploying these systems to report qualifying crashes on a fixed timeline, creating the first centralized, cross-manufacturer dataset regulators and researchers can query.
What actually gets reported
- Any crash where a Level 2 ADAS feature (adaptive cruise, lane centering, etc.) was engaged within 30 seconds of impact and the crash involved a hospital-treated injury, a fatality, a vehicle tow-away, an airbag deployment, or a vulnerable road user
- Any crash involving a Level 3-5 automated driving system while in automated mode, on public roads, regardless of severity — a much lower bar than the Level 2 threshold
- Basic facts: date, location type, weather, whether the system was engaged, and a narrative description
What the data does and doesn't show
The dataset is genuinely useful for spotting emerging patterns — a cluster of similar-sounding incidents pointing to a specific sensor or software condition has triggered investigations before a formal recall was filed. But it's easy to misread. A company operating thousands of automated test vehicles will naturally generate more raw reports than one operating a few dozen, even if its per-mile safety record is better, because the ADS reporting threshold is so low (any crash, not just serious ones). Reading a raw count as a league table without normalizing for fleet size and miles driven is the single most common misuse of this data.
| Comparing | Level 2 ADAS reports | ADS (Level 3-5) reports |
|---|---|---|
| Reporting bar | Higher-severity crashes only | Any crash, any severity |
| Typical fleet size | Millions of consumer vehicles | Hundreds to low thousands of test/service vehicles |
| What a spike usually means | A specific model or software version worth investigating | Needs context on fleet size and miles before conclusions |
A crash report in this database does not establish fault or confirm the system caused the crash. It only confirms the system was active nearby in time — investigation determines cause.
How researchers and journalists use it responsibly
Credible analysis pairs the SGO data with a company's disclosed operational miles or vehicle count to estimate a rough rate, flags when severity thresholds differ between ADAS and ADS reporting, and treats any single cluster as a lead for further investigation rather than a final verdict. NHTSA itself uses the data this way — as an early-warning trigger for formal defect investigations, not as a public rating system.
Where this is headed
Regulators have signaled they want a clearer, more standardized incident-reporting framework as automated driving systems scale toward commercial robotaxi service, including possibly separating reporting tiers more precisely by automation level and adding exposure-mile context. Expect the reporting rules to keep evolving rather than staying fixed, so treat any specific numeric example as a snapshot, not a permanent figure.
Frequently asked questions
Can I look up crash data for a specific ADAS system?
Does a high crash count mean a system is unsafe?
Is this the same data used for recalls?
Do all automated vehicle companies have to report?
How current is the published data?
Sources & further reading
- NHTSA — Standing General Order on Crash Reporting
- NHTSA — Press release on ADAS/ADS crash reporting order
- Federal Register — ADS/Level 2 ADAS incident reporting notice
Figures, prices and policy details were current at the last-updated date above. Automotive pricing, incentives and regulations change frequently — verify time-sensitive details with the linked primary sources. Read our editorial policy and fact-checking standards.
