UL 4600¶
Standard for Safety for the Evaluation of Autonomous Products
Overview¶
UL 4600 is a safety standard for evaluating autonomous products, with a primary focus on self-driving vehicles. It takes a goal-based approach rather than prescribing specific technical solutions, requiring manufacturers to build a comprehensive safety case that demonstrates acceptable risk.
Key Topics¶
- Safety case framework -- structured argumentation showing the system is acceptably safe
- Risk assessment -- identification and evaluation of hazards across the operational design domain
- Dependability claims -- claims about reliability, availability, and integrity of AI components
- Data integrity -- requirements for training data quality, labeling accuracy, and dataset management
- Lifecycle management -- processes for design, testing, deployment, and field monitoring
- Tool qualification -- ensuring development and testing tools are themselves trustworthy
Relevance to vnvspec¶
UL 4600 emphasizes goal-structured notation (GSN) for safety cases. vnvspec's GSN Mermaid exporter (vnvspec.exporters.gsn_mermaid) produces assurance case diagrams directly from specs:
from vnvspec import Requirement
req = Requirement(
id="REQ-SAFE-001",
statement="The system shall achieve ASIL-B integrity for perception outputs.",
verification_method="analysis",
standards={"ul_4600": ["8.1", "8.3"]},
)
The Hazard model's severity, exposure, and controllability fields align with UL 4600's risk assessment methodology.
See also: vnvspec.exporters.gsn_mermaid for GSN assurance case export.