Reviews Report
- Progress is uneven: most consumer vehicles offer supervised SAE Level 2 driver assistance, and NHTSA states no self-driving vehicles are available for purchase. Driverless robotaxis (SAE Level 4) operate only in defined areas with cautious expansion (e.g., California CPUC approvals for Waymo in parts of Los Angeles and San Mateo County) and early safety reports showing lower injury-crash rates within their operating domains (CPUC; Waymo 2024).
- Insurance impacts are being driven more by market-wide cost pressures than autonomy gains: the U.S. CPI for motor vehicle insurance rose roughly 20% year over year in 2024 (BLS CPI). ADAS lowers claim frequency in many cases (IIHS/HLDI), but rising repair complexity and ADAS calibrations increase severity and cycle time, muting net savings (CCC Crash Course 2024; Mitchell ITR).
- Cybersecurity remains a core risk area despite stronger governance: the EU now enforces UN R155/R156 cybersecurity and software‑update requirements for new registrations, OEMs align to UNECE R155/R156 and ISO 24089, and U.S. guidance emphasizes secure design and disclosure (NHTSA). New vulnerabilities continue to surface across vehicles and charging infrastructure (e.g., Pwn2Own Automotive 2024), underscoring the need for continuous hardening.
The idea of a self-driving car was first presented by General Motors at New York’s World Fair in 1939. A long-standing motivation is safety, but the oft-quoted claim that 94% of crashes are due to human error is outdated as a predictor of automation benefits. Evidence indicates that automation will not eliminate most crashes unless it also addresses risky choices like speeding and impairment (IIHS 2020). Recent insurance analyses likewise found no measurable crash-rate reduction from current partial automation packages (HLDI/IIHS 2024). By contrast, targeted automation such as Automatic Emergency Braking is now mandated and is projected to save at least 360 lives and prevent at least 24,000 injuries annually once fully implemented (NHTSA AEB final rule).
Now, most cars contain driver-assistance features such as lane-departure alerts and automatic emergency braking—about 95% of new U.S. light vehicles already include AEB (IIHS). Despite some predictions of widespread autonomy by 2020, NHTSA confirms no consumer-available vehicles today are self-driving. Limited Level 3 availability exists under narrow conditions (e.g., Mercedes‑Benz DRIVE PILOT), while Level 4 robotaxis operate driverless only within defined urban operating domains (Waymo).
Self-Driving Cars Autonomy Levels
The National Highway Traffic Safety Administration (NHTSA) and SAE International reference the SAE J3016 taxonomy as the authoritative framework for driving automation levels (0–5). NHTSA’s public materials use these levels to distinguish driver support (Level 2) from automated driving systems (Levels 3–5), and there have been no new levels added since the April 2021 J3016 revision (SAE J3016; six tiers).
- Level 0: no autonomy. The human performs the full driving task; features may warn or provide momentary intervention, but no sustained automation is present.
- Level 1: driver assistance. The system can assist with either steering or acceleration/braking in limited conditions. The driver supervises and handles fallback.
- Level 2: partial automation. The system can assist with both steering and acceleration/braking simultaneously in limited conditions. Continuous driver supervision and fallback are required.
- Level 3: conditional automation. Within its operational design domain (ODD), the system performs the driving task without continuous supervision but may issue a request to intervene. U.S. availability is limited and narrowly scoped (e.g., traffic-jam scenarios).
- Level 4: high automation. In a defined geofence/ODD, the system performs all driving and fallback; no driver attention is required while active. This underpins today’s driverless robotaxi services in select cities.
- Level 5: full automation. The system can perform all driving under all conditions; no such vehicles are deployed.
How Self-Driving Cars Will Affect Insurance
Fully- or partially-autonomous vehicles, like the Tesla electric vehicle, could change how people drive and their habits. For now, overall premium levels are dominated by macro loss costs: the U.S. CPI index for motor vehicle insurance rose about 20% year over year in 2024 (BLS CPI). While ADAS such as AEB reduce crash and injury claim frequency (IIHS/HLDI), higher repair complexity and post-crash ADAS scanning/calibrations often increase severity and cycle time (CCC; Mitchell), so early expectations that premiums could drastically reduce have not materialized.
Insurers emphasize limited long-run loss experience for higher automation. The NAIC expects any major pricing shifts to be gradual and data-driven, with personal auto remaining central while coverage forms evolve.
Carriers are refining pricing by feature and exposure, integrating VIN-level ADAS data where available and prioritizing telematics/usage-based insurance to capture driving behavior—often a larger rating factor than automation features amid elevated loss costs (CCC; NAIC).
Like cybersecurity markets, initial assumptions are being updated as real-world losses accrue. Repair analytics show hundreds of dollars added to many repairs due to scanning and calibration, along with added days of cycle time, which helps explain why most consumers did not see lower premiums in 2024 despite ADAS benefits (CCC Crash Course 2024; Mitchell ITR).
He adds that it is important to remember that insurance follows the vehicle, not the driver.
Liability allocation will evolve as vehicles assume more of the driving task. The UK’s Automated Vehicles Act 2024 assigns first-party compensation to motor insurers when an authorized self-driving feature is engaged, who may then recover from responsible parties—signaling how insurer-centered compensation could work without immediate premium reductions absent supportive loss data.
In the interim, insurers are taking small steps in altering coverage in response to automated driving. For example, many insurers are now offering discounts to drivers who use in-car monitors and are refining VIN-level pricing credits for specific ADAS features; early Level 4 fleets often use bespoke commercial policies or self-insurance, and if operator-reported safety advantages (e.g., Waymo’s lower injury crash rates) persist at scale, liability costs per mile could decline over time.
Self-Driving Cars Pros and Cons
With a greater focus on the possibility of a self-driving world, it is useful to look at some pros and cons of automated vehicles:
Are Self-Driving Cars Safe?
A 2020 study conducted by the Insurance Institute of Highway Safety (IIHS) cautioned that automated vehicles would not eliminate most crashes unless they also address risky behaviors (e.g., speeding, impairment). Complementing that, insurance claim analyses through 2024 show no measurable crash-rate reduction from current partial automation compared with similar vehicles without those systems.
At the same time, specific active-safety features already deliver measurable benefits. NHTSA’s final rule requiring Automatic Emergency Braking projects at full effectiveness it will save at least 360 lives and prevent at least 24,000 injuries annually (NHTSA). Early Level 4 robotaxi deployments have reported substantially lower police-reportable and injury crash rates than human drivers in the same areas (Waymo 2024), though results are operator-reported and ODD-limited.
- Errors involving “predicting” what another driver will do such as accelerating were not eliminated.
- “Planning and deciding” mistakes such as operating a vehicle at speeds not right for conditions also continued to cause accidents.
- “Execution and performance” errors involving inappropriate maneuvers to avoid a collision or overcompensating in reacting to a situation also remained a cause of accidents.
Not everyone has agreed with the IIHS conclusions. Brad Templeton, who was involved in the self-driving efforts of Google, commented in Forbes that “I think if you asked most self-driving car developers where the hard problems are, they would say that [perception] is the hard one, and [planning and deciding] and [execution and performance] are the easiest to get right.”
