The context — when electronics hold the wheel

A modern car is more than a hundred controllers commanding braking, power steering, transmission, and now driving assistance. When electronics decide, a failure no longer just stalls the engine: it can kill.

The Toyota unintended-acceleration affair, around 2009-2010, crystallised the question — not so much by its actual cause (largely attributed to floor mats and pedals) as by the one it posed to the whole industry: how do you prove an electronic control system is safe? ISO 26262, published in 2011, is the normative answer: adapting functional safety to the automotive world, where volume, cost and usage variability change the rules.

ISO 26262: the IEC 61508 of automotive

ISO 26262 — “Road vehicles — Functional safety” — is the automotive application of IEC 61508. It covers the electrical and electronic (E/E) systems of production vehicles. Its first edition dates from 2011; the second (2018) broadened the scope to trucks, buses and powered two-wheelers, and added a part dedicated to semiconductors.

Like all of the 61508 family, it rests on a safety lifecycle — but with vocabulary and metrics specific to the domain.

ASIL: the scale, determined by risk

Where the process world speaks SIL and machinery PL, automotive speaks ASILAutomotive Safety Integrity Level: QM (quality managed, no safety requirement), then A, B, C, D (the most demanding).

ASIL is not chosen, it is determined by the HARA analysis (Hazard Analysis and Risk Assessment), combining three parameters for each hazardous situation:

  • Severity (S0-S3): from no injury to fatal injuries;
  • Exposure (E0-E4): the probability of being in the situation (rainy motorway, reversing…);
  • Controllability (C0-C3): the driver’s ability to avoid the harm.

Crossing the three gives the ASIL. A severe (S3), frequent (E4) and uncontrollable (C3) situation calls for ASIL D.

The V-model and the safety case

ISO 26262 structures development as a V-model: item definition, HARA, functional safety concept, then technical safety concept, system / hardware / software development, and the ascent through integration, verification and validation.

It is all recorded in a safety case arguing that the item is safe. An automotive specificity: the confirmation measures — confirmation reviews, functional safety audit, and assessment — with increasing degrees of independence depending on the ASIL. The higher the ASIL, the more the assessor must be independent of the design team.

The automotive-specific concepts

Four notions structure the practice:

  • ASIL decomposition: splitting an ASIL D requirement across two independent lower-level elements (e.g., ASIL B(D) + ASIL B(D)), provided their independence is proven. It makes achievable what would be too costly in one block.
  • SEooC (Safety Element out of Context): developing a component without a specific target vehicle, on the basis of usage assumptions, which an integrator then validates. It is the model of Tier-1 suppliers.
  • FTTI (Fault Tolerant Time Interval): the time between a fault appearing and the harm. Fault detection and the transition to a safe state must fit within this window.
  • Hardware metrics: automotive does not measure PFD but coverage rates and an hourly probability.
MetricASIL BASIL CASIL D
SPFM (single-point faults covered)≥ 90 %≥ 97 %≥ 99 %
LFM (latent faults covered)≥ 60 %≥ 80 %≥ 90 %
PMHF (dangerous failure/hour)< 10⁻⁷< 10⁻⁷< 10⁻⁸

The risk reduction to SIL and failure rate to MTBF calculators remain useful for reasoning in orders of magnitude.

SOTIF (ISO 21448): when the function “works” but is not enough

ISO 26262 deals with failures: a component breaks. But ADAS and autonomous driving pose another problem: the system works exactly as designed, and it is the design itself that is insufficient — a camera that fails to recognise a pedestrian against the sun, a radar fooled by a reflection. No failure, but a hazard.

That is the subject of SOTIF — ISO 21448 (Safety Of The Intended Functionality): the safety of the intended function against performance limits and foreseeable misuse. For an autonomous vehicle, SOTIF is at least as critical as ISO 26262: you can no longer reduce everything to a failure. The two standards complement each other.

Cybersecurity: ISO/SAE 21434

A connected vehicle is attackable, and an attack can become a physical hazard. ISO/SAE 21434 is the cybersecurity counterpart of ISO 26262 — Road vehicles — Cybersecurity engineering. Functional safety, SOTIF and cybersecurity now form a triptych inseparable from vehicle development.

The stakes

The human stake is direct: braking, steering, ADAS — failure is measured in lives.

The industrial stake is scale: a systematic defect replicates across millions of vehicles; a recall costs hundreds of millions. Process rigour is not a luxury, it is insurance.

The regulatory stake rises with autonomy: type approval, liability, and a framework (including the EU AI Act for AI functions) that is tightening.

The mistakes not to make

MistakeWhy people make itWhy it is dangerous
Believing ISO 26262 covers ADAS on its ownIt is THE automotive safety standard26262 handles failures, not functional insufficiency — you need SOTIF (21448)
Decomposing into ASIL without proving independenceCutting the cost of an ASIL DWithout real independence, a common cause ruins the decomposition
Using a SEooC without checking the assumptionsReusing an “already safe” componentThe supplier’s usage assumptions may not hold in your vehicle
Forgetting the FTTIDetecting the fault… too lateDetection slower than the tolerance window does not protect
Separating safety and cybersecurityTwo teams, two worldsA cyber flaw can trigger the very hazard functional safety aimed to prevent

Good practice

Run a rigorous HARA per situation, and derive the ASIL — do not “choose” it by habit.

Treat ISO 26262 (failures), SOTIF (functional insufficiency) and ISO/SAE 21434 (cyber) together, especially for ADAS and autonomy.

Justify any ASIL decomposition with demonstrated independence.

Verify a SEooC’s assumptions in the real integration context.

Size fault detection and the transition to a safe state to fit within the FTTI.

Document a reasoned safety case and organise confirmation measures with the independence the ASIL requires.

Review checklist

  • Item definition clear and bounded
  • HARA documented (S, E, C) and ASIL determined per function
  • Functional then technical safety concepts traced
  • ASIL decomposition justified by proven independence
  • SEooC assumptions verified at integration level
  • Hardware metrics met (SPFM, LFM, PMHF) for the target ASIL
  • FTTI covered by detection + safe state
  • SOTIF (ISO 21448) addressed for ADAS/autonomous functions
  • Cybersecurity (ISO/SAE 21434) integrated
  • Safety case and confirmation measures with adequate independence

Going further

One last thing. The automotive industry pushed functional safety to mass scale, then forced it to surpass itself: with autonomy, proving a system does not fail is no longer enough — you must prove that, even working perfectly, it does the right thing. That is the whole leap from ISO 26262 to SOTIF, and the real challenge of the decade.