StandardGuidancev1.0.0

ISO/IEC 24029 series: assessment of the robustness of neural networks

The international series making neural network robustness testable, usable as methodology under Article 15 while prEN 18229-2 remains in draft.

The official source remains authoritative. This is general information about obligations and not legal advice. See this object on the map

Address and citation

This object has an address of its own that is never renamed or reused. Store the identifier in your own file, not the title or the link.

Identifier: praxikon:eu:ai-act:standard:standard-iso-iec-24029-robustness-neural-networks
Payload hash (sha256): 90a88e1eab0ea2115fc69dcffb95fd81f4beda2674e8a39e98349b2cddc54e5e

Citation line

Praxikon, "ISO/IEC 24029 series: assessment of the robustness of neural networks", praxikon:eu:ai-act:standard:standard-iso-iec-24029-robustness-neural-networks@1.0.0, dataset praxikon:sys:registry:dataset:ai-act-implementation-graph 2.1.0 (schema 1.4.0), effective_at 2026-08-08T00:00:00.000Z, known_at 2026-08-14T00:00:00.000Z, sha256 90a88e1eab0ea2115fc69dcffb95fd81f4beda2674e8a39e98349b2cddc54e5e

Version: 1.0.0
Legal time (effective_at): 8 August 2026
Knowledge time (known_at): 8 August 2026
Closed on: Not closed
Topics: standards

Review status: Placed against the official source (8 August 2026). Next check due by 4 February 2027. The check date is the knowledge date of this version; no later recheck has been recorded.

What this object links to

Every relation appears below as a path: from the source with its locator, through the conditions and exceptions of the object carrying the relation, to the consequence. A locator belongs to a statement in the data and not to a relation, so the source is the source anchor of the carrying object.

The obligation this hangs off

1 of 1 shown

The object belongs to this obligation. The source line it hangs off sits there.

Source
Official fact on this object, with its locator.
- ISO/IEC JTC 1/SC 42: international standards for artificial intelligence
  Locator: ISO/IEC TR 24029-1:2021 and ISO/IEC 24029-2:2023, ISO/IEC JTC 1/SC 42
  praxikon:eu:ai-act:source:iso-iec-jtc1-sc42
  Open official source
- EU Artificial Intelligence Act 2024/1689
  Locator: Article 15
  praxikon:eu:ai-act:source:reg-eu-2024-1689
  Open official source
Via
- Condition | allParticularly relevant to high-risk systems based on neural networks.
Consequence
- ObligationArticle 15: accuracy, robustness and cybersecurity
  praxikon:eu:ai-act:obligation:article-15-accuracy-robustness

What this object is about

1 of 1 shown

The object is about this role. Undifferentiated: it does not follow that the duty rests on this role.

Source
Official fact on this object, with its locator.
- ISO/IEC JTC 1/SC 42: international standards for artificial intelligence
  Locator: ISO/IEC TR 24029-1:2021 and ISO/IEC 24029-2:2023, ISO/IEC JTC 1/SC 42
  praxikon:eu:ai-act:source:iso-iec-jtc1-sc42
  Open official source
- EU Artificial Intelligence Act 2024/1689
  Locator: Article 15
  praxikon:eu:ai-act:source:reg-eu-2024-1689
  Open official source
Via
- Condition | allParticularly relevant to high-risk systems based on neural networks.
Consequence
- ActorProvider of an AI system
  praxikon:eu:ai-act:actor:provider

What this object states

Official fact

Attributable to a named primary source, with a locator. Where they differ, the official source prevails.

ISO/IEC TR 24029-1:2021 (Assessment of the robustness of neural networks, Part 1: Overview) is a technical report mapping the topic and available assessment methods. ISO/IEC 24029-2:2023 (Part 2: Methodology for the use of formal methods) describes the application of formal methods in assessing robustness. The series is aimed at AI developers and users assessing robustness across the lifecycle. Neither part is cited in the Official Journal, so no presumption of conformity under Article 40 arises. For Article 15 the designated deliverable is prEN 18229-2, which as at June 2026 was still in drafting.
- Locator: ISO/IEC TR 24029-1:2021 and ISO/IEC 24029-2:2023, ISO/IEC JTC 1/SC 42praxikon:eu:ai-act:source:iso-iec-jtc1-sc42Open official source
- Locator: Article 15praxikon:eu:ai-act:source:reg-eu-2024-1689Open official source

Our interpretation

Our own reading. It can change without the law changing, and it is not the position of a supervisory authority.

Part 1 is a technical report and therefore not a requirements set: it gives an overview, not a benchmark. Part 2 is the sharper text substantively, but formal methods are only feasible for relatively bounded models. For large generative models you hit the limits quickly, leaving empirical testing. The trap is making a robustness claim measured only on the average test set: Article 15 is specifically about behaviour under errors, faults and unexpected input.
- Locator: ISO/IEC TR 24029-1:2021 and ISO/IEC 24029-2:2023, ISO/IEC JTC 1/SC 42praxikon:eu:ai-act:source:iso-iec-jtc1-sc42Open official source
- Locator: Article 15praxikon:eu:ai-act:source:reg-eu-2024-1689Open official source

Recommended step

A practical step we consider appropriate. Not an obligation following from the Regulation.

Choose deliberately per system between formal verification (for bounded models) and empirical stress testing, and record the choice and its justification. Test explicitly on out-of-distribution input, missing fields and edge cases, and document the performance degradation observed. Add those results to the technical documentation and to the instructions for use where they define the limits of use.
- Locator: ISO/IEC TR 24029-1:2021 and ISO/IEC 24029-2:2023, ISO/IEC JTC 1/SC 42praxikon:eu:ai-act:source:iso-iec-jtc1-sc42Open official source
- Locator: Article 15praxikon:eu:ai-act:source:reg-eu-2024-1689Open official source

When this applies

1Particularly relevant to high-risk systems based on neural networks.

When this does not apply

No exception recorded on this object.

Referring to this object

Citation block

Copy this reference into your advice, article or file. The identifier, the version and the hash keep the statement findable later, even once the dataset has moved on.

Reference

Praxikon, "ISO/IEC 24029 series: assessment of the robustness of neural networks",
praxikon:eu:ai-act:standard:standard-iso-iec-24029-robustness-neural-networks@1.0.0,
dataset praxikon:sys:registry:dataset:ai-act-implementation-graph 2.1.0 (schema 1.4.0),
effective_at 2026-08-08T00:00:00.000Z, known_at 2026-08-08T00:00:00.000Z,
sha256 90a88e1eab0ea2115fc69dcffb95fd81f4beda2674e8a39e98349b2cddc54e5e,
https://www.praxikon.com/en/verkenner/standard/standard-iso-iec-24029-robustness-neural-networks
(https://www.praxikon.com/api/v1/entities?id=praxikon%3Aeu%3Aai-act%3Astandard%3Astandard-iso-iec-24029-robustness-neural-networks&effective_at=2026-08-08&known_at=2026-08-08&lang=en, accessed 2026-08-25)

Short form

praxikon:eu:ai-act:standard:standard-iso-iec-24029-robustness-neural-networks@1.0.0 (sha256 90a88e1e)

BibTeX

@misc{praxikon-eu-ai-act-standard-standard-iso-iec-24029-robustness-neural-networks-1-0-0,
  author       = {{Praxikon}},
  title        = {ISO/IEC 24029 series: assessment of the robustness of neural networks},
  year         = {2026},
  version      = {1.0.0},
  number       = {praxikon:eu:ai-act:standard:standard-iso-iec-24029-robustness-neural-networks},
  howpublished = {AI Act Change \& Evidence Graph, dataset 2.1.0, schema 1.4.0},
  note         = {effective_at 2026-08-08T00:00:00.000Z; known_at 2026-08-08T00:00:00.000Z; sha256 90a88e1eab0ea2115fc69dcffb95fd81f4beda2674e8a39e98349b2cddc54e5e},
  url          = {https://www.praxikon.com/en/verkenner/standard/standard-iso-iec-24029-robustness-neural-networks},
  urldate      = {2026-08-25},
  language     = {en}
}

CSL JSON

[
  {
    "id": "praxikon:eu:ai-act:standard:standard-iso-iec-24029-robustness-neural-networks@1.0.0",
    "type": "dataset",
    "title": "ISO/IEC 24029 series: assessment of the robustness of neural networks",
    "container-title": "AI Act Change & Evidence Graph",
    "publisher": "Praxikon",
    "version": "1.0.0",
    "number": "praxikon:eu:ai-act:standard:standard-iso-iec-24029-robustness-neural-networks",
    "URL": "https://www.praxikon.com/en/verkenner/standard/standard-iso-iec-24029-robustness-neural-networks",
    "language": "en",
    "issued": {
      "date-parts": [
        [
          2026,
          8,
          8
        ]
      ]
    },
    "accessed": {
      "date-parts": [
        [
          2026,
          8,
          25
        ]
      ]
    },
    "note": "dataset praxikon:sys:registry:dataset:ai-act-implementation-graph 2.1.0; schema 1.4.0; effective_at 2026-08-08T00:00:00.000Z; known_at 2026-08-08T00:00:00.000Z; sha256 90a88e1eab0ea2115fc69dcffb95fd81f4beda2674e8a39e98349b2cddc54e5e; retrieved_from https://www.praxikon.com/api/v1/entities?id=praxikon%3Aeu%3Aai-act%3Astandard%3Astandard-iso-iec-24029-robustness-neural-networks&effective_at=2026-08-08&known_at=2026-08-08&lang=en; licence https://www.praxikon.com/nl/legal/terms"
  }
]

How to verify a reference later is set out in the methodology. Terms

For agents and integrations

This page and the machine output come from the same object and the same two time axes.

Open as JSON Open as JSON-LD Read the methodology