Contents
StandardISO

ISO 21448 (SOTIF)

Road vehicles — Safety of the intended functionality

ISO 21448 helps engineers address unsafe vehicle behavior caused by limitations in a function’s design or performance, even when nothing has failed.

What it is

ISO 21448 addresses safety of the intended functionality, usually shortened to SOTIF. It concerns unreasonable risk arising from insufficient functional specifications or performance limitations in their implementation. A function can operate as designed and still behave unsafely in a situation its designers did not adequately address.

For example, a perception function might struggle to distinguish an obstacle against a confusing background without any component having failed. SOTIF asks whether that limitation can lead to hazardous vehicle behavior and how the risk is addressed.

Who it applies to

The standard concerns intended functions implemented through E/E systems in series production road vehicles, excluding mopeds. It is especially relevant where safe behavior depends on situational awareness obtained through complex sensors and algorithms, including driver assistance and automated driving. Reasonably foreseeable misuse is within scope. These scope boundaries are described in the ISO catalogue entry.

For project planning, vehicle manufacturers and suppliers need clear responsibilities for assumptions, interfaces and evidence. A perception supplier may characterize detection limits, while the vehicle team assesses their consequences for braking, steering and driver interaction.

What it covers

The standard provides a framework for arguing that the intended functionality is sufficiently safe, supported by design, verification, validation and operational activities.

A practical way to organize the engineering questions is:

Question Example work product
What should the function do, and under which conditions? Function description and operating assumptions
Where could its capabilities be insufficient? Analysis of limitations and triggering conditions
How could those limitations create hazardous behavior? Vehicle scenarios and risk assessment
How will the risk be reduced and evaluated? Mitigation requirements and validation evidence

For a camera-based assistance function, useful investigations might explore glare, partial occlusion, unusual objects and driver expectations. These are examples to assess against the actual function, rather than a universal test checklist.

How it fits with other documents

ISO 21448 complements ISO 26262: functional insufficiencies and malfunctioning behavior require different analysis, although their consequences can overlap. For example, poor detection under a difficult scene and detection lost through a sensor fault may both affect braking, but need different supporting evidence.

Teams should connect their analyses so that assumptions and mitigations remain consistent. An ASIL assignment alone does not establish that a perception function handles its intended operating situations adequately.

Cybersecurity threats are outside SOTIF’s scope. Coordination with cybersecurity engineering remains useful where shared interfaces or assumptions influence vehicle behavior. SOTIF also contributes to AI safety work, but it does not replace detailed evaluation of a particular model and its integration.

In practice

Start with a precise feature description: operating conditions, driver responsibilities, limitations and behavior when capability becomes insufficient. Turn these assumptions into reviewable requirements and scenarios.

Combine complementary evidence such as analysis, simulation, controlled testing and appropriately planned vehicle testing. Explain why the evidence supports the safety argument; a large distance driven or a good average detection score needs context to be meaningful.

On an HPC application processor, review the complete path from sensing through perception and decisions to vehicle action. Revisit the argument when software, models, calibration or operating assumptions change, and use field observations to investigate previously overlooked situations.

What it means for your discipline

  • Functional safetyCore

    Own the SOTIF safety argument, hazard analysis, acceptance criteria and supporting evidence. Coordinate investigation of hazardous behavior that can occur without a fault.

  • HPC & application softwareCore

    Identify limitations in perception, prediction and decision algorithms, then implement and validate mitigations. Treat software that meets its specification as potentially insufficient for some real driving situations.

  • E/E architectureCore

    Define function boundaries, operating assumptions and interactions with the driver and other systems. Allocate sensing, monitoring and fallback capabilities that address identified limitations.

  • Hardware & siliconPartial

    Provide realistic sensor and processing performance limits across relevant operating conditions. Help distinguish normal capability limits from hardware faults.

  • CybersecurityPartial

    Coordinate analyses when similar unsafe outcomes could arise from functional limitations or malicious interference. Keep cybersecurity threat evidence distinct from the SOTIF argument.

  • Software update & OTAPartial

    Assess whether changes to models, calibration or feature behavior invalidate existing SOTIF evidence. Support targeted revalidation before release.

No direct stake: Base software · Networks & communication · Diagnostics & lifecycle