Autonomous driving systems (ADS) represent a transformative shift in mobility, with the potential to dramatically reduce traffic accidents caused by human error. According to established classifications, SAE J3016 defines six levels of driving automation. Level 0 involves no automation, with the human driver fully responsible. Level 1 provides driver assistance for either steering or acceleration/deceleration. Level 2 offers partial automation, such as adaptive cruise control combined with lane centering, but the driver must remain attentive and ready to intervene. Level 3 introduces conditional automation, where the system handles the dynamic driving task in specific conditions, allowing the human to disengage but requiring readiness for fallback. Level 4 achieves high automation, operating without human intervention within a defined operational design domain (ODD), such as geofenced areas or specific conditions. Level 5 delivers full automation, capable of driving anywhere under all conditions without any human input or controls.
In practice as of February 2026
Most deployed systems remain at Level 2, with widespread adoption of features like adaptive cruise control and lane-keeping in consumer vehicles. Level 3 deployments have been limited and sometimes paused due to regulatory and liability complexities, while Level 4 operations thrive in controlled environments, such as robotaxi services in urban areas with defined routes and weather constraints. These higher levels promise significant safety gains by eliminating human factors like distraction or fatigue, which contribute to a large percentage of crashes.
Achieving true safety in ADS requires adherence to multiple complementary standards. ISO 26262 focuses on functional safety, addressing risks from random hardware failures and systematic errors in electrical/electronic systems through Automotive Safety Integrity Levels (ASIL). It mandates rigorous processes like hazard analysis and risk assessment (HARA), fault tree analysis, and verification to ensure systems respond safely to malfunctions. For autonomous features, updates in ongoing editions incorporate considerations for automated driving systems, emphasizing fail-operational designs where critical functions continue despite faults.
ISO 21448, known as Safety of the Intended Functionality (SOTIF), complements ISO 26262 by targeting hazards that arise even when the system functions as designed. These include performance limitations (e.g., sensor degradation in fog), foreseeable misuse (e.g., unexpected driver behavior in fallback scenarios), and functional insufficiencies in complex environments. SOTIF requires identifying triggering conditions—scenarios that expose weaknesses—and mitigating them through design refinements, extensive validation, and ongoing monitoring. As of 2026, revisions and applications of ISO 21448 continue to evolve, with emphasis on scenario-based approaches for Level 3+ systems.
Cybersecurity integration via ISO/SAE 21434 is equally critical, as connected ADS are vulnerable to attacks that could compromise safety. This standard demands threat modeling, secure development lifecycle practices, and resilience against intrusions.
Regulatory landscapes are advancing rapidly.
In the United States, the NHTSA’s Automated Vehicle Framework, advanced under the current administration, prioritizes ongoing safety of public road operations, innovation by reducing unnecessary barriers, and enabling commercial deployment. Recent actions include modernizing Federal Motor Vehicle Safety Standards (FMVSS) for vehicles without traditional controls, such as updating requirements for transmission, defrosting, and lighting to accommodate ADS. The proposed SELF DRIVE Act of 2026 aims to formalize a national framework, strengthening NHTSA authority, improving crash data transparency, and promoting U.S. leadership in AV technology while avoiding a patchwork of state regulations. NHTSA’s ongoing rulemakings and public meetings, including those planned for March 2026, focus on exemptions, incident reporting, and safety assessments.
Globally, the United Nations Economic Commission for Europe (UNECE) has made strides with the adoption of a draft Global Technical Regulation (GTR) on Automated Driving Systems in early 2026 by the Working Party on Automated/Autonomous and Connected Vehicles (GRVA). This builds on prior frameworks, facilitating safe introduction of unsupervised self-driving vehicles on public roads worldwide, with provisions for performance, assessment, and data collection. Expected formalization later in 2026, it aligns with UN regulations like UN-R 157 for automated lane-keeping systems.
Building a robust safety case is fundamental.
This comprehensive documentation demonstrates how the ADS safely performs the dynamic driving task, incorporating hazard identification, risk mitigation strategies, safety management systems, and continuous performance monitoring. It includes evidence from design, verification, validation, and real-world operations.
Best practices center on layered validation.
Extensive simulation is essential, generating millions of scenarios—including rare edge cases like adverse weather, construction zones, erratic pedestrians, or sensor occlusions—to bridge coverage gaps that real-world testing alone cannot achieve. High-fidelity tools model realistic physics, sensor behaviors (lidar, radar, cameras), and traffic interactions, while addressing the sim-to-real gap through domain adaptation and hybrid testing. Hardware-in-the-loop and vehicle-in-the-loop setups further validate integrated systems.
Real-world piloting follows in controlled ODDs, starting small and expanding based on data. Scenario-based testing per SOTIF identifies triggering conditions, with ethical AI design ensuring transparent, predictable decision-making in dilemmas (e.g., prioritizing vulnerable road users). Redundancy in sensors, computing, and actuators enhances fault tolerance.
Phased development supports this: Begin with precise requirements definition aligned to standards, incorporate redundancies in architecture, conduct iterative verification through analysis and testing, and maintain post-deployment updates via over-the-air mechanisms and fleet learning for continuous improvement.
This holistic approach integrating standards, rigorous testing, regulatory alignment, and adaptive processes transforms ADS from promising technology into proven, reliable systems. In 2026’s landscape, where deployment accelerates amid policy advancements, prioritizing safety not only minimizes risks but also builds public trust, accelerates adoption, reduces crashes, and unlocks benefits like enhanced mobility for all.
