Allianz Calls for EU Driving License to Certify Autonomous Vehicles

• Edge-case reliability and unexpected scenarios: Long-tail events (rare weather, unusual roadworks, unconventional human behavior) remain hard to predict and validate. Failures outside the vehicle’s operational design domain can cause catastrophic accidents and erode public trust. Known unknowns: completeness of scenario coverage, generalization beyond training/test data, and robustness under degraded sensors. Impact: [Fatal]

• Cybersecurity with broadened data access: Opening “accident- and safety-related in-vehicle data” and maintaining connected fleets expands the attack surface. Risks include remote control, sensor spoofing, malware in perception/decision pipelines, and exfiltration of sensitive trip/telemetry data. Known unknowns: EU-wide security standards for the proposed database, secure data minimization, identity/authentication across stakeholders, and supply-chain integrity. Impact: [Fatal]

• Systemic software/OTA defects causing correlated failures: Homogeneous stacks and simultaneous over-the-air updates can propagate a single defect across fleets and borders, creating synchronized incidents and overwhelming emergency and insurance systems. Known unknowns: rigor of pre-deployment verification, rollback guarantees, fleet-level kill‐switch governance, and cross-border recall coordination. Impact: [Fatal]

• Certification adequacy and “teaching to the test”: A unified “driving license” could still underrepresent real-world complexity if tests overweight set-piece maneuvers and simulations. Vendors may optimize to pass criteria without true robustness. Known unknowns: scenario diversity targets, near‐miss reporting thresholds, performance under compounding failures (e.g., sensor blockage + adverse weather), and how learning from incidents updates certifications. Impact: [Critical]

• Liability alignment and cross-border consistency: Strict owner liability may misalign incentives if vendor or supplier faults dominate, and fragmented national implementations could yield inconsistent claims handling and dispute resolution. Catastrophic events could stress insurer solvency and premium affordability. Known unknowns: apportionment among OEMs, software providers, owners, and infrastructure operators; harmonized evidentiary standards for data logs; and reinsurance capacity for tail risks. Impact: [Critical]

• Nov 10, 2025 — Abu Dhabi Autonomous Summit (Microsoft, WeRide, Archer Aviation): Expect announcements on city-scale robotaxi and drone regulatory sandboxes, MOUs with municipal regulators, and 2026 pilot launch timelines; any commitments to publish safety KPIs or near-miss data would set a precedent relevant to the EU’s push for standardized testing and transparency in autonomous mobility.

• Next 6–12 months — EU decision on an AV “driving license” framework: Watch for a European Commission Call for Evidence and scoping under DG MOVE/CNECT that could kick off an impact assessment for an EU-wide license regime for autonomous systems, catalyzed by Allianz’s data (e.g., 66% fewer reversing crashes with ADAS; 30% drop in rear-end collisions with AEB) and its forecasted 20% accident reduction by 2035.

• Early 2026 — Draft pathways to embed the “license” model into EU type-approval: Monitor for proposals to amend or complement Regulation 2022/1426 to codify a certification stack combining standardized simulation suites, closed-course maneuvers (emergency braking, evasive maneuvers), and real-world trials (night, motorway, mixed traffic), including formal Operational Design Domain (ODD) declarations and performance thresholds aligned with Level 3–4 capabilities.

• 2026 — EU data-access mandate and near-miss database blueprint: Expect a Commission recommendation or delegated act clarifying what counts as “accident- and safety-related in-vehicle data,” including retention, anonymization, and access rights for manufacturers, insurers, and regulators; look for the initial governance proposal for a joint European database of critical traffic situations and near misses to enable proactive safety oversight and model validation.

• 2026 — Insurance and liability alignment for AVs: Track consultations or guidance confirming strict owner liability regardless of human vs autonomous control mode, plus updates addressing rising repair costs for sensor/compute-heavy vehicles; catalysts include insurer-led pilot pricing frameworks for Level 3–4 coverage and potential supervisory feedback on claims handling, subrogation, and fault apportionment in mixed human–autonomy incidents.

• H1 2026 — Public trust and transparency milestones: Watch for new pan-EU sentiment tracking (e.g., a Eurobarometer module or large-scale insurer surveys) focusing on reliability in edge cases, alongside proposed transparency measures such as publication of anonymized incident/near-miss data or standardized safety dashboards; movement here will directly address concerns that 69% worry about unexpected situations and 72% view the tech as “too new.”

• 2026 global harmonization checkpoints — UNECE WP.29 and cross-border recognition: Follow automated/connected vehicle sessions for updates that could dovetail with an EU “AV license,” such as harmonized test protocols, data-logging requirements, or ODD labeling—foundational for mutual recognition of certifications across markets and for reducing fragmentation that currently slows deployment and complicates liability frameworks.

• 2026–2027 — Member-state pilots and pre-standard trials: Anticipate national authorities to launch pilot certification programs reflecting the license concept (simulation + track + on-road) as a prelude to EU-level adoption; key catalysts include public reporting on safety outcomes vs human driving baselines and integration of ADAS success metrics (e.g., AEB-driven reductions) into acceptance criteria, which will influence legislative drafting speed and insurer pricing models.

Autonomous “driving licenses” sound tidy, but the proposal risks conflating human-centered metaphors with software realities. Allianz’s data on ADAS gains is promising, yet it does not settle the harder questions of generalization, governance, and incentives in Level 3–4 autonomy. A certification that mimics human testing might produce checklists and theater rather than true system safety—especially once over‐the‐air updates change behavior after approval.

Questions 1) Is the “driving license” analogy fit for software? Who holds the license—the model, the vehicle type, the fleet software version, or the operator—and how is it revoked or renewed after each update, domain shift, or learned behavior change? 2) Do ADAS outcomes predict L3–L4 risk? How will risk compensation, mixed traffic effects, rare tail events, cyberattacks, and edge-case generalization be measured beyond simple crash-rate deltas? 3) Who should bear strict liability? If owners are liable for machine decisions they cannot foresee or control, will this suppress consumer adoption and push the market into insurer-approved fleet oligopolies?

Different perspectives and alternative views

• Safety engineers: License regimes risk ossifying around test suites that are easy to pass and easy to game; continuous post-deployment evidence and hazard discovery matter more than pre-deployment rituals.
• Privacy advocates: “Open safety data” can become de facto mass telematics surveillance; purpose limitation, minimization, and strong differential privacy are table stakes.
• Small OEMs and startups: EU-wide licensing may entrench incumbents with the resources to pass exhaustive tests and maintain compliance infrastructure.
• Municipal planners: More lives may be saved sooner by mandating ISA, default AEB, urban speed limits, and road redesign than by waiting for L3–L4 rollout.
• Labor and equity groups: Fleets could erase driver jobs while concentrating mobility control; public interest obligations and access pricing should be defined ex-ante.
• Cybersecurity experts: Centralized, “open” event databases become high-value targets; safety and attack surface must be co-governed.

Controversial or provoking statements

• If AVs are demonstrably safer, the ethical endgame is to restrict human driving on motorways by a set date; anything less renders “safety first” a slogan.
• If insurers demand AV data access, reciprocity should apply: open loss data, pricing models, and claims decisions to public audit to prevent risk selection and hidden discrimination.
• Strict owner liability for autonomous operation is a moral hazard; product liability should shift primarily to manufacturers and software providers.
• A license without post-deployment auditability is safety theater; continuous, standardized, privacy-preserving telemetry plus randomized audits should be mandatory.
• EU competitiveness may suffer if certification cycles lag software iteration; the faster path to safety could be hard caps on speed, universal AEB, and road design—then autonomy.