Page 26 - The Annual AI Governance Report 2025 Steering the Future of AI

The Annual AI Governance Report 2025: Steering the Future of AI

committee is simultaneously mapping ISO/IEC texts to EU regulatory clauses and developing
new metrics (e.g., trustworthiness indicators for autonomous vehicles) to secure cross-border
interoperability within the single market.

4.2 Technical Standards Development Standards Theme 4: AI

Types of Technical Standards: AI standards fall into four practical categories. Management-
system standards, such as ISO/IEC 42001 , require organisations to establish roles, risk
73
processes, and continuous improvement loops for each stage of AI development. A similar
approach is adopted at the network level by the ITU-T Y.3061 architecture for autonomous
telecom networks. Assessment and assurance standards address individual systems. ISO/
74
IEC 42005 establishes an eight-step method for impact assessment , while ITU-T Y.3173
75
provides metrics for rating the 'intelligence' of 5G nodes and services. Conformity and audit
76
standards lay the foundations for third-party certification. ISO/IEC 42006 sets out the rules for
the competence of auditors of AI management systems , while ITU-T F.781.2 specifies the
77
test protocols and evidence thresholds for AI-driven medical software. At the interface layer,
78
technical implementation standards such as ITU-T Y.3172 define the data formats and control
points that enable machine-learning functions to be integrated into 5G networks. Meanwhile,
79
socio-technical practice guides such as IEEE 7000 provide engineers with guidance on value
elicitation and traceability, ensuring that ethical considerations are incorporated into software
development. 80

Gaps in Current Technical Standard Setting: Adoption of formal AI standards remains the
exception rather than the rule as firms are faced with a patchwork of faster—but unofficial—
industry frameworks, resulting in inconsistency and forum-shopping risks. Participation is
81
likewise skewed: civil-society groups, SMEs and many Global-South delegations lack the
money and time to engage, while Big Tech staff attend up to 80 hours a week, giving them
disproportionate influence over committee outcomes. Traditional standards development
82
procedures, which often take 18–36 months from proposal to publication, cannot keep pace
with the rapid evolution of agentic AI models. Such procedures also struggle to codify value-
83
laden issues, such as fairness, across diverse legal regimes. A recent study by the Oxford Martin
School thus argues that traditional Standard Development Organisations should shift their focus

73 ISO/IEC 42001:2023. (2023). ISO
74 ITU-T Recommendation database. (2023, December 14). ITU-T Y.3061. ITU.
75 ISO/IEC. (2025) 42005:2025. ISO.
76 OECD.AI. (2024, July 2). ITU-T Y.3173 - Framework for evaluating intelligence levels of future networks
including IMT-2020.
77 ISO (2025). ISO/IEC 42006.
78 ITU. (2024, June 13). F.781.2- Quality assessment requirements for artificial intelligence/machine learning-
based software as a medical device.
79 ITU. (2019, June 22). Y.3172: Architectural framework for machine learning in future networks including
IMT-2020.
80 IEEE Standard. (2021, September 15). 7000-2021 - IEEE Standard Model Process for Addressing Ethical
Concerns during System Design | IEEE Xplore.
81 Huw R. and Ziosi M. (2025, June 9) Can we standardise the frontier of AI? Oxford Martin AI Governance
Initiative.
82 Huw R. and Ziosi M. (2025, June 9) Can we standardise the frontier of AI? Oxford Martin AI Governance
Initiative. Page 9.
83 Huw R. and Ziosi M. (2025, June 9) Can we standardise the frontier of AI? Oxford Martin AI Governance
Initiative. Page 10.

17