Chapter 2 · EV08

Safety Certification Rules Misaligned with AV Operating Reality

Date collected: 26 May 2026

Summary

Traditional safety certification frameworks predate AV sensor-fusion, edge-case autonomy; misalignment creates regulatory gaps, stalling deployment and accountability

EV08 — Safety Certification Rules Misaligned with AV Operating Reality

Summary for Drafting

Current safety certification frameworks (NHTSA FMVSS, UK CAVS/AVA, Euro NCAP) were designed around human driver assumptions and are demonstrably misaligned with how autonomous systems actually operate. The exemption process exists precisely because of this mismatch—yet exemption processes are slow, uncertain, and create deployment delays. Cases include: Cruise’s failed exemption petition (2018-2020, no action), Waymo’s workaround compliance approach (using pre-existing FMVSS rather than ADS certification), and the UK’s new “self-driving test” framework (still being defined). This evidence supports the “Institutional ODD” argument: governance frameworks need to codify what operating conditions systems are certified for, and align certification with actual deployment scenarios, not theoretical human-equivalence.

Key Data Points

NHTSA FMVSS Exemption Process

Exemptions Granted (as of 2026):

Only one company (Nuro Inc.) has received a general exemption from FMVSS requirements (low-speed autonomous vehicles)
General Motors / Cruise: Petitioned NHTSA in 2018 to allow deployment of driverless Chevrolet Bolt AVs (no steering wheel, no pedals) on U.S. roads. Petition withdrawn in 2020 after years of no action. Company later faced regulatory issues and ceased operations.

Process Timeline Issue:

Historically: Exemption application processing took multiple years
2025 Reform: NHTSA streamlined Part 555 exemption process, anticipating reduction from years to months for processing
Bottleneck: Part 555 allows only 2,500 vehicles/year for exempt production—a significant constraint for scale-up

Governance Lesson: The slowness of exemptions creates a catch-22: manufacturers cannot deploy to test certification claims, certification bodies have no real-world data to validate, and policy uncertainty discourages investment in compliance pathways.

UK CAVS Act / Automated Vehicles Act 2024

Certification Framework:

Vehicles must pass a “self-driving test” to be authorised as “automated vehicles”
The test must ensure vehicles “travel autonomously and can do so safely and legally”
Detailed specifications of the self-driving test have not yet been defined (as of February 2026)

Exemption Route:

Trialling organizations unable to comply can apply to the Vehicle Certification Agency (VCA) for a Vehicle Special Order (VSO) under s.44 Road Traffic Act 1988
VSO process requires a “reasoned case” for exemption
Secretary of State approval required

Deployment Timeline:

Automated Passenger Services (APS) regulations acceleration planned to facilitate commercial pilots from Spring 2026 with paying passengers and no safety driver
Implementation Programme published 26 February 2025

Governance Lesson: The UK framework explicitly acknowledges that existing regulations don’t fit AV operating reality (hence the exemption route), and is building a new institutional ODD structure. However, the “self-driving test” is still undefined, creating ongoing uncertainty for operators.

Euro NCAP AV Testing Protocols

Assessment Structure (2026 onwards): Euro NCAP moved away from L3/L4-specific testing toward a “Balance Principle” assessing:

Vehicle Assistance (speed assistance, steering assistance, Adaptive Cruise Control performance)
Driver Engagement (driver monitoring, system status, driving collaboration)

2026 Protocol Revisions:

Four Stages of Safety: Safe Driving → Crash Avoidance → Crash Protection → Post-Crash Safety
Introduction of robustness testing with real-world scenario variations (speed, offset) to test unpredictability
Automation of 1,200+ tests under updated protocols

Gap with Operating Reality:

Euro NCAP does NOT test full L4 autonomous systems—focuses on L2 assistance and “assisted driving”
Protocols emphasize preventing “over-reliance on automation” (assumes human driver in loop)
Real L3/L4 systems operate without assumption of constant driver engagement
Governance Lesson: Testing standards still lag actual deployment scenarios. Real-world autonomy requires different safety paradigms than assisted driving.

Evidence Tier

Type: Policy gap analysis + limited case study (Cruise exemption) Strength: Medium-high (clear documentation of exemption process, timeline delays, and framework gaps; limited quantitative data on real-world misalignment incidents)

Specific Cases & Citations

Case 1: Cruise NHTSA Exemption — Failed Application & Deployment Collapse

Timeline: Exemption petition filed 2018 → Withdrawn 2020 (no action) → October 2023 crash incident dragged pedestrian 20 feet → October 2023 California DMV suspended Cruise license → NHTSA closed investigation August 2024
Governance Failure: Exemption process neither approved nor denied; company operated under regulatory uncertainty → safety incident → enforcement
Citation:
- NHTSA press release: “NHTSA Announces Consent Order with Cruise After Company Failed to Fully Report Crash Involving Pedestrian” (2023)
- NHTSA Equipment Recall Report: “RMISC-23E086-4326” (2023)
- Telematics Wire: “NHTSA closes inquiry of GM’s Cruise robotaxis” (August 2024)
- TechCrunch: “Teamsters urge NHTSA to deny Cruise Origin exemption” (September 2023)

Case 2: Waymo Compliance Workaround — FMVSS Pre-Certification Rather Than ADS Certification

Approach: Rather than seeking autonomous driving system (ADS) certification, Waymo uses dual-mode vehicles (steering wheel + autonomous capability) certified by Original Equipment Manufacturer (FCA for Pacifica minivans) under existing FMVSS
Misalignment: FMVSS requirements assume human driver control; Waymo circumvents this by keeping steering wheels that are not actually used
Safety Performance: Swiss Re analysis shows 92% fewer bodily injury claims and 88% fewer property damage claims over 25 million miles vs. human drivers
Governance Lesson: Company solved misalignment by NOT seeking direct ADS approval; instead retrofits existing compliant vehicles. This works for Waymo but doesn’t scale to driverless-only designs.
Citation:
- Butzel Long: “WAYMO BECOMES THE FIRST TO START TESTING LEVEL 4 VEHICLES ON PUBLIC ROADS WITHOUT A DRIVER” (source: Waymo safety documentation)
- Waymo Safety Case Approach White Paper
- Waymo Research: “Comparison of Waymo Rider-Only crash rates by crash type to human benchmarks at 56.7 million miles” (Journal of Transportation Safety & Security, 2025)

Case 3: NHTSA Part 555 Reform — Acknowledgment of Processing Delay Problem

Issue: Exemptions historically took years to process, causing manufacturers to abandon applications or seek workarounds
Reform (2025): NHTSA streamlined Part 555 process under Trump/Duffy Administration with goal of reducing processing from years to months
Remaining Constraint: Part 555 still limits exempt vehicles to 2,500/year
Governance Lesson: Exemption process bottleneck was recognized as impediment to innovation; structural reforms attempted but volume ceiling remains
Citation:
- NHTSA press release: “U.S. Transportation Secretary Sean P. Duffy Streamlines Exemption Process for Noncompliant Automated Vehicles” (2025)
- Covington & Burling: “NHTSA Proposes New Autonomous Vehicle Program” (January 2025)
- WardsAuto: “NHTSA revises exemption process for vehicles without driver controls”

Additional Context: Emerging Standardization Efforts

UL 4600 / UL 4740 Standards:

UL 4600 addresses “Autonomous Vehicle Safety Standard” for L4/L5 vehicles, extending ISO 26262 (functional safety) and ISO 21448 (functional safety of systems not designed by humans)
Explicitly defines “Operational Design Domain” (ODD) as critical compliance dimension
UL 4740 addresses sensor resilience (e.g., lidar contamination, dirt, vehicle fluids)
Governance Lesson: Private standards bodies (UL) are filling gaps left by public regulation, but fragmentation remains.

Citation:

UL Solutions: “Autonomous Vehicle Safety Training and Advisory”
APEX.ai: “Legislation, regulations, and standards for automated mobility” (comprehensive taxonomy overview)

Drafting Notes for Chapter 2 (“What Is CAM?”)

Use this evidence to argue:

Current Institutional Gap: Safety frameworks (FMVSS, CAVS Act, Euro NCAP) are not adequately calibrated to autonomous operating reality. FMVSS assumes human driver; CAVS Act codifies “self-driving test” but test is undefined; Euro NCAP focuses on assisted driving, not autonomous operation.
Exemption as Evidence of Mismatch: The fact that exemptions exist (Nuro, Waymo workaround, Cruise attempted) proves the gap. Exemption processes are slow and uncertain precisely because they’re bridge mechanisms between misaligned regimes.
The Institutional ODD Argument: Governance must explicitly codify what operating conditions systems are certified for (ODD), and align certification with actual deployment scenarios. The UK AVA is moving in this direction with “self-driving test,” but details matter.
Scale-Up Constraint: Part 555 exemption ceiling of 2,500 vehicles/year prevents mass deployment. Modernization of FMVSS (via proposed 2026 rules, AV STARTUP Act, SELF DRIVE Act) is necessary but political uncertainty remains.

Key Quote for Chapter: “Safety certification rules were written for human drivers. Autonomous systems operate differently. Exemptions exist precisely because of this mismatch—but exemptions are slow, uncertain, and limit scale-up. Governance must define operational design domains explicitly and align certification with real deployment scenarios.”

Sources

Thread