ISO 21448: SOTIF for AI/ML Components — Insufficiencies of Learned Functions and the ISO 8800 Bridge

Date
2026-10-05
Location
Online
Host
Zach L
Register

About this event

A live 30-minute expert session on SOTIF for AI/ML Components — Insufficiencies of Learned Functions and the ISO 8800 Bridge (ISO 21448 — Safety of the Intended Functionality). What We'll Cover: Why ML components generate SOTIF-type insufficiencies without faults and where they enter the ISO 21448 analysis The method step by step: training-data ODD coverage review, robustness and OOD testing, uncertainty calibration checks, and triggering-condition pairing per failure mode How ML evidence maps to ISO 21448 Clause 7 and 10 and aligns with ISO/PAS 8800 AI-safety lifecycle artifacts Common mistakes: accuracy metrics presented as safety evidence, test sets drawn from training distribution only, no retraining impact analysis The traceability auditors want: dataset-to-ODD mapping, per-failure-mode test evidence, and a controlled retraining and re-validation process Related topics: machine learning safety · iso 8800 · distribution shift · out of distribution detection · neural network robustness · uncertainty calibration · training data coverage · ml insufficiency · ai in adas · dataset odd mapping · retraining impact · learned function validation · ml failure modes · safety of ai Critical Systems Analysis provides embedded functional safety consulting for ISO 21448 — Safety of the Intended Functionality. Learn more: https://criticalsystemsanalysis.com Partner with us: https://meetings.hubspot.com/benjamin-twombly/strategic-partnerships

Topics

About Functional Safety Foundations

This online session delivers expert-led instruction on the regulatory landscape and implementation approach for Functional Safety Foundations.

The cross-sector fundamentals underlying every functional safety standard: risk as severity × likelihood, the safety function concept, integrity levels, the safety lifecycle, and the split between random and systematic failures.

What it regulates

Shared vocabulary and reasoning: hazard → risk → risk reduction allocation → safety requirement → verified implementation → maintained integrity in operation.

Who must comply

Everyone entering the field — the conceptual base of the entire catalog.

Key requirements of Functional Safety Foundations

  • Hazard and risk fundamentals
  • Safety function anatomy: sensor-logic-actuator with defined safe state
  • Integrity ladders (SIL/PL/ASIL/DAL) as risk-reduction currencies
  • Random vs systematic failure treatment
  • Lifecycle and management of functional safety incl. competence

Key concepts: Functional Safety Foundations

Risk reduction allocation
Distributing required risk reduction among mechanical design, safeguards, control functions and procedures.
Functional safety management
Planning, competence, documentation and assessment wrapped around technical work.
Random vs systematic
Hardware wear-out/chance failures quantified statistically versus design/process errors controlled by rigor.
Safe state
The defined condition a function drives the equipment to on demand or fault.

Frequently asked questions: Functional Safety Foundations

Why can't testing alone prove safety?

Rare-event targets (like 1e-7/h) are unverifiable by test duration, and systematic errors evade random testing — hence the standards' dual machinery of quantified hardware analysis plus process rigor and independent assessment.

Standard information based on the published text of Functional Safety Foundations. Event content is provided by the host. For authoritative guidance, consult the official standard body.