ISO/IEC 24029-2:2023
ISO/IEC's methodology for proving the robustness of neural networks with formal methods: define the domain by bounded attributes, express stability, sensitivity, relevance and reachability criteria, choose solvers, abstract interpretation, reachability analysis or model checking to fit the network, and apply them from design through verification, deployment and operation.
ISO/IEC 24029-2:2023 is a compliance framework from International with 3 domains and 18 controls. The largest domains are Clause 5: Robustness properties and criteria (domain, stability, sensitivity, relevance, reachability) – ISO/IEC 24029-2:2023 (8 controls), Clause 7: Robustness assessment through the AI system life cycle – ISO/IEC 24029-2:2023 (7 controls), Clause 6: Choosing formal methods that fit the network and its data – ISO/IEC 24029-2:2023 (3 controls). Every control below carries what it requires and what an assessor expects to see.
Framework summaries on this platform are interpretations grounded in the published source, for educational and compliance planning purposes. They do not reproduce or replace the official standards. Refer to the authoritative source for the definitive text. Framework names and trademarks belong to their respective organisations.
Framework Domains (3)
Clause 5: Robustness properties and criteria (domain, stability, sensitivity, relevance, reachability) – ISO/IEC 24029-2:2023
| Code | Title |
|---|---|
| iso-iec-24029-2-2023::5.2 | Specify the domain of use through bounded attributes |
| iso-iec-24029-2-2023::5.3.1 | Express a stability property where noisy inputs must not change the output |
| iso-iec-24029-2-2023::5.3.2 | Define and apply a stability criterion |
| iso-iec-24029-2-2023::5.4.1 | Carry out sensitivity analysis over a domain |
| iso-iec-24029-2-2023::5.4.2 | Define and apply a sensitivity criterion |
| iso-iec-24029-2-2023::5.5.1 | Use a relevance property where a human could do the task |
| iso-iec-24029-2-2023::5.5.2 | Define and apply a relevance criterion |
| iso-iec-24029-2-2023::5.6 | Express and satisfy a reachability criterion for closed-loop agents |
Clause 6: Choosing formal methods that fit the network and its data – ISO/IEC 24029-2:2023
| Code | Title |
|---|---|
| iso-iec-24029-2-2023::6.1 | Match the formal method to the network architecture and input data type |
| iso-iec-24029-2-2023::6.2.1 | Classify and choose verifiers by completeness, determinism, model access and arithmetic |
| iso-iec-24029-2-2023::6.2.6 | Check every model when model checking is used |
Clause 7: Robustness assessment through the AI system life cycle – ISO/IEC 24029-2:2023
| Code | Title |
|---|---|
| iso-iec-24029-2-2023::7.2.2 | Identify the features the network has learned with a relevance criterion |
| iso-iec-24029-2-2023::7.2.3 | Check class separability with a sensitivity criterion |
| iso-iec-24029-2-2023::7.3.2 | Bound and justify the parts of the input domain covered at verification |
| iso-iec-24029-2-2023::7.3.3 | Measure the impact of perturbations formally |
| iso-iec-24029-2-2023::7.4 | Check the effect of numerical accuracy at deployment |
| iso-iec-24029-2-2023::7.5.2 | Monitor robustness against the planned domain of operation |
| iso-iec-24029-2-2023::7.5.3 | Assess changes in robustness when the network keeps learning |
What is ISO/IEC 24029-2:2023 and who does it apply to?
ISO/IEC 24029-2:2023 is a compliance framework from International with 3 domains and 18 controls. ISO/IEC's methodology for proving the robustness of neural networks with formal methods: define the domain by bounded attributes, express stability, sensitivity, relevance and reachability criteria, choose solvers, abstract interpretation, reachability analysis or model checking to fit the network, and apply them from design through verification, deployment and operation. It is used by organisations to establish and maintain compliance with industry standards and regulatory requirements.
What does ISO/IEC 24029-2:2023 actually require?
ISO/IEC 24029-2:2023 has 18 controls organised across 3 domains. The largest domains are Clause 5: Robustness properties and criteria (domain, stability, sensitivity, relevance, reachability) – ISO/IEC 24029-2:2023 (8 controls), Clause 7: Robustness assessment through the AI system life cycle – ISO/IEC 24029-2:2023 (7 controls), Clause 6: Choosing formal methods that fit the network and its data – ISO/IEC 24029-2:2023 (3 controls). Each control defines specific requirements that organisations must implement to achieve compliance.
If I already comply with another framework, how much of ISO/IEC 24029-2:2023 do I already cover?
ISO/IEC 24029-2:2023 does not currently have cross-framework mappings in our system. Check back as we continuously expand our mapping database.
How do I implement ISO/IEC 24029-2:2023?
Start your ISO/IEC 24029-2:2023 compliance journey by running a self-assessment on our platform to identify your current compliance posture. Our AI advisory can answer specific questions about ISO/IEC 24029-2:2023 requirements, and cross-framework mapping helps you leverage existing controls from other frameworks you may already comply with. Create a free account to access all 18 controls and track your progress.
Start Your Compliance Journey
Create a free account to run self-assessments, get AI advisory, and track your compliance progress across 1,248 frameworks.
Get Started Free →Free forever, no credit card required