Section 3 · Assessing AI governance

Exploring who is responsible for overseeing and evaluating AI

The previous section showed that meaningful evaluation must follow an AI-assisted finding from the original question and evidence through to interpretation, use and consequence.

The project is now focused on understanding more about the governance context in different countries. We are currently interviewing experts across the UK and Ireland on the governance of AI: when it should be used, how it should be used, who is responsible for overseeing and evaluating it, and what should happen when something goes wrong.

AI governance documents are growing, but what do they say about evaluation?
0 20 40 60 80 2021 2022 2023 2024 2025 2026 UK-wide Republic of Ireland Scotland to June

The growing number of documents shows increasing attention to AI governance. It does not tell us whether those documents require meaningful evaluation, identify who should oversee it, or explain what must happen when a finding fails.

Across the UK and Ireland we are analysing governance documents and interviewing people who hold different positions in the system. That work is ongoing, so no comparative findings are reported here.

Why institutional context matters

Approaches to AI governance are more complex than a simple choice between regulation and innovation. Different systems distribute authority, responsibility and rights of challenge in different ways, and a duty on paper does not by itself create a route through which an affected person can challenge a finding or secure correction.

This is why situated evaluation is part of the governance question. Whether a finding is adequate depends on the institution using it, the decision it informs, the people it affects and the consequences of error.

The aim is not to rank countries or governance systems. It is to understand whether evaluation is connected to oversight, responsibility, challenge and action in practice.

Have your say: what should oversight require?

What to take from this

Testing can identify a problem. Governance determines who must respond, who can challenge and whether anything changes.

What this shows

The framework and questions guiding our ongoing analysis of AI evaluation and governance across the UK and Ireland.

How to read it

This is a research programme, not a set of findings. The growing number of governance documents shows increasing activity, not the strength or effectiveness of oversight.

What this cannot tell us

Which governance system provides the strongest oversight or whether formal commitments work in practice. The document analysis and interviews are ongoing.

Why this matters

Evaluation protects nobody unless it is connected to oversight, responsibility, challenge, correction and remedy.