Page 53 - AI Standards for Global Impact: From Governance to Action

AI Standards for Global Impact: From Governance to Action

7�1�2 Group 2: Standards, Best Practices and Conformity Assessment

This group focused on technical and methodological aspects of AI testing collaboration:

Priorities and gaps for collaboration Part 2: Thematic AI

a) Technical aspects of model testing including testing environment, data requirements, and
reproducibility standards
b) Comprehensive risk assessment covering misuse risks, AI-cyber intersections, AI-bio risks,
and broader socio-technical challenges

Approaches to close the gaps

a) Development of technical reports on standards mapping for trustworthy AI testing
b) Establishment of pre-standardization discussions, potentially leveraging ITU platforms (e.g
ITU-T Focus Groups).

7�1�3 Group 3: Institutional frameworks

The institutional frameworks group examined governance and coordination mechanisms:
Key institutional gaps

a) Need for identifying areas requiring global alignment in AI testing approaches
b) Requirements for agile governance structures that can adapt quickly to technological
developments
c) Strategic information-sharing, knowledge-building, and co-production mechanisms are
currently inadequate

Solutions and coordination mechanisms

a) Establishing comprehensive coordination frameworks with effective feedback loops
b) Emphasizing complementary roles among international organizations like ITU while
avoiding duplication of efforts

7�2 Future directions

Where do we go from here? Participants agreed to continue the dialogue on collaboration for
trustworthy AI testing initiated at the AI for Good Global Summit to discuss how to enact some
of the proposals made. Some of the key actions proposed were:

a) Continued dialogue on trustworthy AI testing: Establishing a regular dialogue among
the various stakeholders involved in the event was emphasized as an important need in
the space, with Group 2 highlighting the need for dialogue focused on frontier model
security testing and Group 1 emphasizing the need for collaboration on capacity building
to enable AI testing across different jurisdictions.
b) Develop technical reports: Working towards technical reports on topics related to
trustworthy AI testing, such as testing environments, protocols, and risk management
frameworks, was highlighted as a valuable next step. Group 3 raised the importance of
strategic information-sharing, knowledge-building, and co-production mechanisms for
greater institutional capacity around the world.

41