Autonomous vehicles: type-approval versus self-certification
Lesson 5 of 5 in Sector by Sector: Health, Employment, Finance, Education, and Vehicles.
Automated driving is the sector where two philosophies of safety regulation meet head-on — and where the EU AI Act deliberately steps aside.
The shared vocabulary is SAE J3016: six levels of driving automation. Levels 0–2 are driver-assistance — the human drives, always (Tesla’s “Full Self-Driving (Supervised)” is, legally, Level 2). Level 3 is conditional automation: the system drives within its design domain and the human is the fallback on request. Level 4 needs no human fallback within its domain (Waymo’s robotaxis); Level 5 — everywhere, all conditions — exists nowhere. The regulatory cliff is between 2 and 3: at Level 3 responsibility migrates from driver to system, and with it, liability and the entire evidentiary burden of safety.
Europe regulates through type-approval: a vehicle system must be approved before sale against UNECE regulations. UN R157 (2021) was the first binding international rule for Level 3 — Automated Lane Keeping Systems, initially capped at 60 km/h and extended to 130 km/h with lane changes in 2022; Mercedes-Benz’s Drive Pilot became the first internationally type-approved Level 3 system under it. UN R155 and R156 add mandatory cybersecurity and software-update management systems — every over-the-air update to a driving function happens inside an approved process. The EU’s implementing regulation 2022/1426 extends type-approval to full automated driving systems (Level 4), and Germany’s 2021 Autonomous Driving Act created the first national framework for Level 4 operation with a remote technical supervisor instead of a safety driver.
The United States inverts the model: self-certification. Manufacturers certify compliance with FMVSS (none of which yet addresses automated driving logic); NHTSA polices afterwards through investigations, recalls, and its Standing General Order requiring crash reports for ADS and Level 2 systems. Congress has failed for a decade to pass an AV statute (SELF DRIVE Act, AV START), leaving a state patchwork — California alone requires DMV permits, disengagement reports, and CPUC authority for paid rides.
| Dimension | EU/UNECE type-approval | US self-certification |
|---|---|---|
When safety is checked | Before market: approval against UN R157/R155/R156, EU 2022/1426 for Level 4 | After market: NHTSA investigations, Standing General Order crash data, recalls |
Who bears the initial burden | Manufacturer proves conformity to an approval authority | Manufacturer self-certifies against FMVSS; burden shifts to NHTSA to prove a defect |
Software updates | Inside an approved software-update management system (UN R156) | OTA updates are free — but can themselves constitute the “remedy” in a recall (Tesla 2023) |
AI Act interplay | Vehicles sit in Annex I Section B: AI Act high-risk requirements flow in via future type-approval delegated acts, not directly | No equivalent; state AI statutes generally carve out vehicle-safety systems |
Key terms: sae levels, type approval, self certification, odd, software update management
Interactive sorting exercise: Place each system at its SAE level — the classification that decides who is legally driving.
Tool: Global Governance Atlas — Map these five sectors onto the jurisdictions you have studied — see which regulator reaches which system in the Governance Atlas.
Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.