Page 64 - The Annual AI Governance Report 2025 Steering the Future of AI

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

as positive examples. The consensus was that future governance frameworks must embed
participation from the outset, not treat it as an afterthought.

Quote: Pillars Chapter 2: Ten

• "The code of practice ... was led by independent chairs and co-chairs with strong
multi-stakeholder support. We have had more than 1000 stakeholders involved in
this and ... this multi-stakeholder aspect is very important." (Juha Heikkilä, Adviser
for Artificial Intelligence, European Commission)

Dive deeper in the Whitepaper “Themes and Trends in AI Governance”:
• 3.3 Regional AI partnerships
• Annex – Examples of multilateral initiatives [a list of some 40 initiatives]
• Annex – Examples of national initiatives [a list of some 20 initiatives]

2.3 Transparency as a Cornerstone of Trust

Transparency is seen by many as paramount to gain public trust. Yet the reality, panelists noted,
is that transparency has regressed even as models grow more powerful. Documentation of
training data, model limitations, and evaluation benchmarks has become thinner in recent
releases, leaving policymakers, researchers, and the public in the dark.

Professor Robert Trager (Co-director, Oxford Martin AI Governance Institute, University of
Oxford) lead a series of discussions throughout the AI of Good Global Summit on how to best
address the challenges of AI verification, i.e., the process by which one party can check or
validate the actions or assertions of another. The goals of these discussions were to identify
gaps in the current AI testing ecosystem and explore solutions. Topics covered included:

• Capacity building for testing AI systems worldwide.
• Developing best practices and standards.
• Creating institutional frameworks for international collaboration.

Concrete proposals included mandatory model cards, registries of AI systems, watermarking
of AI-generated content, and disclosure of intended uses and known risks. Several participants
stressed that transparency must extend beyond the technical level: governments and companies
should be clear about how decisions are made, who is accountable, and how citizens can
challenge harmful outcomes.

The process of moving from research to pre-standardization and eventually to official
standardization is necessary but challenging due to the rapid evolution of AI. The hurdles in
verifying AI for trustworthiness are significant.

55