Each part of the sentence gives us somewhere to look

We do not have settled answers to every governance question. What evidence should be required before deployment? Which uses should be prohibited? How should responsibility be divided between a model developer and an organization using its model?

Inquiry means investigating those questions, comparing alternatives, and testing assumptions. Practice means someone must turn an answer into a procurement clause, an evaluation, a reporting requirement, or a decision to suspend a system.

For the ministry, inquiry might establish whether the triage tool performs reliably across different patient groups. Practice decides who conducts that evaluation, what evidence the ministry receives, and what result would stop the purchase.

EvidenceAllan Dafoe’s 2018 research agenda approached AI governance through questions about the technical landscape, political dynamics, and desirable governance arrangements: an early case for treating governance as a subject of research.2

Research and implementation need each other. A mechanism built on a poor understanding of the problem may enforce the wrong thing. An insight that never reaches a decision may leave the problem untouched.

Governments have authority. Companies control products and infrastructure. Researchers possess expertise. Workers understand how systems change their jobs. Affected people know consequences that may never appear in a performance report.

These roles carry different responsibilities and different claims to influence. Expertise can help determine whether a tool works. It cannot, by itself, settle which trade-offs people should accept.

The ministry may have the authority to buy the tool. Clinicians and patients still need ways to influence how it is used: information, accessible channels, and some route through which an objection can change a decision. A meeting alone does not provide that.

EvidenceThe UN’s 2024 Governing AI for Humanity report found that 118 countries were parties to none of the seven prominent non-UN international governance initiatives it examined. That is a defined sample rather than every global discussion, but it exposes a substantial representation gap.3

That is why the definition says societies. The plural keeps differences of authority, interest, and circumstance in view. Decisions made in one place can affect people elsewhere, and their interests belong in the question even when their governments or communities have little say in the answer.

Governance can restrict an unsafe use. It can also fund research, train public officials, improve access to computing resources, and support locally relevant development.

For the ministry, governing the tool could mean requiring evidence before purchase. It could equally mean investing in the staff needed to evaluate suppliers intelligently in the first place.

EvidenceThe African Union’s Continental AI Strategy shows this enabling side: its agenda includes skills, research and innovation, and an institutional environment for applications in health, agriculture, and education.4

A useful tool may never reach the people who need it. A government may stay dependent on services it cannot understand, negotiate over, or replace. Those are governance concerns too, alongside harmful deployment.

Imagine two companies that both believe further safety testing would be useful. Testing takes three months. Each believes its competitor will release in six weeks. Their intentions matter. So does the pressure to move first.

EvidenceThat pressure appears in published policy. OpenAI’s 2025 Preparedness Framework update said it “may adjust” its requirements if a rival released a high-risk system without comparable safeguards, subject to stated conditions. It shows how competition enters governance decisions; it does not establish that any particular release was unsafe.5

Governance asks what would make careful testing a decision companies can sustain. Buyers might require evidence. Disclosure requirements might make skipped testing visible. Liability might change who bears the cost of failure. Each idea needs scrutiny of its own; the starting point is to understand what behavior the environment encourages.

The same logic runs inside the ministry. If a department is rewarded only for processing patients faster, it may give too little attention to mistakes, accessibility, or unequal performance. If reporting problems threatens a team’s budget, monitoring may produce reassurance rather than useful evidence.

Institutions give authority and responsibility somewhere to live: regulators, courts, company boards, procurement offices, professional bodies, and independent evaluators.

For the ministry: who approves the purchase? Who can inspect the evidence? Who investigates a complaint? Who can suspend use?

Naming an organization is only the beginning. It needs expertise, resources, access to information, and enough independence to use its authority. A regulator may have a strong mandate and too few staff. A company’s safety team may understand a problem and lack the power to delay a release. A complaints office may receive objections without being able to secure a remedy.

EvidenceJennifer Wang and colleagues scored what sixteen companies publicly disclosed about implementing their voluntary White House commitments. The average was 53%; the highest-scoring company reached 83%. Public disclosure is not the same as internal practice, and that is the point: without ways to obtain evidence, outsiders cannot tell whether a commitment was kept.6

AI relies on computing resources, data, networks, electricity, technical skills, and capital. How those are distributed shapes who can build, who can adapt, and who must accept someone else’s terms.

EvidenceThe World Bank frames AI readiness around connectivity, compute, context, and competency, and reports that high-income countries hosted 77% of global co-location data-center capacity in mid-2025.7

The ministry might buy a tool that performs well but runs on a proprietary service. Over time, patient records, staff training, and clinical workflows become tied to it, while the supplier controls prices and product changes.

Foreign supply is not automatically a governance failure. Cooperation and specialization can provide valuable capabilities. The practical questions are about information, bargaining power, data portability, and the ability to replace a service. A seat at the table does not create those abilities by itself. They require capacity.

“AI should be accountable” expresses a principle. A safeguard makes some part of that principle operational. For the ministry, that could mean testing on relevant patient populations, protecting patient data, recording recommendations, letting clinicians override them, investigating incidents, and giving people a route to challenge consequential decisions.

Five questions show how fully a safeguard has been specified:

1. 1Who acts?
2. 2What is required?
3. 3What evidence is produced?
4. 4What decision follows?
5. 5What happens if the requirement is not met?

Suppose the tool performs poorly for one patient group. Who detects that? Does the evidence trigger further testing, a restriction, or suspension? And who verifies that the response actually happened?

Human review deserves particular care. A person’s presence does not guarantee meaningful oversight. They need time, relevant information, competence, and the authority to intervene.

EvidenceAustralia’s Robodebt Royal Commission called for clear review routes, plain-language explanations of automated decisions, independent scrutiny of algorithms, and a body empowered to monitor and audit automation. Robodebt concerned automated administration rather than today’s generative AI; its lesson is about making scrutiny and remedy possible.8

Power is the ability to set terms, obtain information, refuse a deployment, demand an explanation, or secure a remedy.

Governance mechanisms create power too. Imagine a requirement for independent evaluation before certain systems can be deployed. If evaluation is expensive or offered by only a few providers, established companies may find it easier to comply, and a protective measure could raise barriers for smaller organizations.

EvidenceGirish Sastry and colleagues make a related warning about compute governance: interventions through computing resources may help achieve safety and other goals, while poorly scoped approaches can threaten privacy or centralize power.9

That calls for better design rather than no design: proportionate requirements, credible evaluators, and attention to who can access the process.

A system can improve a headline measure while spreading its gains and costs unevenly.

EvidenceA widely used healthcare algorithm predicted spending as a stand-in for medical need. Because less had been spent on Black patients, it underestimated their needs: at the same risk score, Black patients were sicker than White patients. Correcting the disparity would raise the share of Black patients selected for additional help from 17.7% to 46.5%.10

That was a care-management algorithm, not a triage tool, but the lesson transfers: accurately predicting a chosen target can coexist with an unfair allocation of help.

So the ministry has to decide what improvement means. Shorter waiting times? Better access? Fewer missed urgent cases? Better working conditions? Which groups should benefit, and what evidence would show that they do?

EvidenceUNESCO’s Recommendation commits to human rights, dignity, fairness, inclusion, and accessible benefits: a standard to pursue, rather than proof that any particular policy meets it.11

Two verbs, because no amount of testing can remove every uncertainty before a system goes live. Preventing harm is work done in advance. Addressing it is work done for the people already affected.

Switching off a faulty tool does nothing for the patients it has already misjudged. Addressing harm means responding to the person, and to the arrangement that let it happen.

Some things should not be done to people, even when someone claims a large overall benefit. Rights set limits that a simple calculation cannot override.

Later, it may be choosing whether to renew a service its hospitals have reorganized around. Staff have been trained. Data formats have changed. Older procedures have disappeared. The decision is still reversible in principle. In practice, reversal has become expensive and disruptive.

EvidenceDavid Collingridge described this dilemma in 1980: early on, a technology’s consequences are hard to foresee; later, when they are clear, changing it may be much harder. “When change is easy, the need for it cannot be foreseen.”12

That leaves two questions to hold together. Do we have enough evidence to act? What happens to our ability to act if we wait?

Preserving choice does not mean keeping every option open forever. A justified prohibition deliberately closes one. The aim is to make consequential commitments through legitimate, informed decisions, and keep ways to correct the ones that fail.

That is why the definition says “as”. It describes a process already under way, not a deadline after which governance becomes pointless.

A rule can be wrong. An evaluation can miss something important. A reporting requirement can impose costs while producing little useful information.

Governance also becomes embedded. Agencies acquire mandates. Organizations build compliance systems. People develop interests in keeping things as they are. So revision has to be part of the original design.

For the ministry, learning depends on information: performance monitoring, incident reports, complaints, independent evaluation, and records that let investigators understand what happened. An institution cannot respond to consequences it cannot see.

EvidenceStilgoe, Owen and Macnaghten’s responsible-innovation framework joins anticipation with reflexivity, inclusion, and responsiveness. In plain language: consider consequences, examine your assumptions, involve others, and be able to respond.13

Adaptation needs accountability as well. Who can request a review? What evidence triggers it? Who can revise the requirements, and must they explain why? Can affected people challenge the change?