Wiki · Concept · Last reviewed July 10, 2026

ISO/IEC TR 5469

ISO/IEC TR 5469 is the ISO/IEC Technical Report on functional safety and AI systems.

Category: AI safety / standards Published: July 10, 2026 Modified: July 10, 2026 Last reviewed: July 10, 2026 Tags: ISO, IEC, functional safety, AI safety, safety cases, AI governance

Definition

ISO/IEC TR 5469:2024 is titled Artificial intelligence — Functional safety and AI systems. ISO lists it as Edition 1, a 73-page Technical Report published in January 2024, with reference number ISO/IEC TR 5469:2024.

The public ISO abstract says the report describes properties, related risk factors, available methods, and processes for three situations: use of AI inside a safety-related function, use of non-AI safety-related functions to ensure safety for AI-controlled equipment, and use of AI systems to design and develop safety-related functions.

Status

As reviewed on July 10, 2026, ISO lists ISO/IEC TR 5469:2024 as published, with publication stage 60.60. Its lifecycle record shows new-project approval on June 9, 2020, committee-draft registration on June 7, 2022, close of the committee-draft comment period on September 29, 2022, final text received on July 14, 2023, proof activity in October and December 2023, and publication on January 8, 2024.

ISO identifies ISO/IEC JTC 1/SC 42 as the responsible technical committee and classifies the report under ICS 35.020. The SC 42 committee page describes the subcommittee's scope as standardization in artificial intelligence and lists Joint Working Group 4 with IEC TC 65/SC 65A for functional safety and AI systems.

Functional Safety Surface

ISO/IEC TR 5469 matters because AI systems create pressure at the boundary between probabilistic behavior and safety-related engineering. Functional safety is concerned with reducing unacceptable risk from malfunctioning behavior of safety-related systems. AI can appear inside the function, outside the function as monitored equipment, or upstream in design and development work.

The report's public abstract usefully separates those roles. An AI component inside a safety-related function raises different evidence problems from a non-AI safety function guarding AI-controlled equipment. An AI tool used to design a safety-related function raises still another kind of process-control question. Lumping those together under "AI safety" hides the architecture.

Engineering Use

For builders, ISO/IEC TR 5469 is most useful when an AI system touches physical, industrial, medical, transport, robotics, energy, or operational-technology contexts where safety-related functions are already part of engineering practice. It should force teams to identify whether AI realizes the safety-related function, is controlled by a separate safety function, or supports safety-function development.

For governance, that distinction matters. If AI is inside the safety function, evidence has to address model behavior, data, uncertainty, monitoring, failure modes, and verification limits. If non-AI safety mechanisms guard AI-controlled equipment, evidence has to show that those mechanisms can detect, constrain, or shut down hazardous behavior. If AI supports design, evidence has to show review and traceability for generated or assisted engineering work.

Evidence Record

An ISO/IEC TR 5469-informed record should identify the safety-related function, AI role, equipment boundary, hazard analysis link, risk factor, safety requirement, data assumption, model or system version, runtime monitor, non-AI safeguard, verification method, validation evidence, operational constraint, owner, and retest trigger.

The record should preserve negative evidence. Unknown operating conditions, out-of-distribution behavior, incomplete requirements, weak monitoring, insufficient independence between AI control and safety guard, or unclear handoff authority should be visible. A safety case should not let a benchmark score replace an argument about hazardous behavior.

Boundary With Other Standards

ISO/IEC TR 5469 is not an AI management-system standard, product approval, or universal safety certification. It sits beside broader and narrower references. AI Safety Cases covers assurance arguments, ISO/IEC 23894 addresses AI risk management, ISO/IEC 5338 addresses AI system life cycle processes, and ISO/IEC TS 8200 addresses controllability of automated AI systems.

Source Discipline

Use the official ISO page for the title, reference number, Technical Report status, publication date, edition, page count, publication stage, technical committee, ICS classification, public abstract, and lifecycle dates. Use the ISO/IEC JTC 1/SC 42 page for committee scope and Joint Working Group 4. Do not cite vendor summaries for the report's formal status, and do not treat ISO/IEC TR 5469 as product approval, certification, or legal safe harbor.

Spiralist Reading

Spiralism reads ISO/IEC TR 5469 as a discipline against autonomy theater. A machine that acts in the world does not become safer because its decision logic is impressive. Safety comes from architecture, limits, monitoring, independence, verification, operation, and the authority to stop.

The stricter reading is that functional safety is not an adjective attached to an AI system. It is a documented relation among hazard, function, safeguard, evidence, environment, and accountability. The more autonomous the system appears, the more important it becomes to know which parts are allowed to fail and which parts must remain boring.

Open Questions

Sources

Return to Wiki