AI procurement in assessment is not just a buying decision; it is a validity, accountability, and public trust decision. The central question is whether an AI tool improves assessment outcomes enough to justify its costs, dependencies, and risks. Stronger sources point towards treating procurement as a governance exercise that needs local evidence, clear accountability, and an exit plan. The open question is how much independent proof is enough before AI is allowed to influence assessment operations or decisions.

AI procurement, governance, and public value is the question of how assessment bodies buy and oversee AI in ways that improve outcomes without importing hidden cost, weak evidence, or avoidable dependency. In assessment, the issue is not only whether a tool works technically. It is whether the organisation can justify the purchase in terms of validity, fairness, accountability, privacy, accessibility, and value for learners and the public.

Assessment leaders are often asked to treat AI as a solution to workload, scale, or modernisation. That can be sensible, but it can also obscure what the system is buying: a new dependency, a new data flow, a new support burden, and a new governance obligation. Public value matters because assessment systems are part of qualification credibility, learner trust, and sometimes public regulation.

- **Public value**: whether the use of AI serves legitimate educational or regulatory goals, not just efficiency. - **Validation**: evidence that a tool works in the specific assessment context where it will be used. - **Responsible procurement**: buying with clear evidence requirements, accountability, and exit planning. - **Material cost**: the hidden financial, environmental, labour, privacy, or organisational costs of deploying AI. - **Governance fit**: whether policy, contracts, oversight, and appeal routes match the risk of the use case.

The stronger source set points in the same direction: procurement is not a neutral buying exercise, but an assessment design decision with governance consequences. The Australian framework is useful because it frames AI in schools as something that needs ethical and responsible use, not open-ended enthusiasm. Helen Beetham’s critique pushes further by asking whether public-sector AI is genuinely effective and whether institutions are being asked to adopt systems before the evidence is strong enough. Crawford’s *Atlas of AI* adds a material lens: AI can carry extraction, infrastructure, and dependency costs that matter when institutions rely on opaque vendor systems and cloud services. The U.S. Education Department’s toolkit reinforces the same practical idea from a different angle: the AI system should be designed around educational purpose, safety, privacy, and equity, not simply capability. There is also broad convergence that governance has to start before deployment. Assessment teams need evidence thresholds, contract clarity, accountability for errors, and a plan for withdrawal or change if the tool underperforms. The public-value lens helps keep efficiency claims in proportion by asking who benefits, who bears the risk, and what is lost if the AI system is adopted badly.

The open question is how much evidence is enough to justify AI procurement in assessment. Vendor and policy documents often assume that responsible adoption is a matter of good intentions and guardrails, but the harder issue is comparative proof: compared with what, in which setting, and at what cost? Helen Beetham’s piece is useful precisely because it asks whether public-sector AI is being oversold as practical help while remaining unproven in real use. Another contested issue is the balance between centralised standards and local fit. A broad framework can help, but assessment bodies still need to decide what evidence is relevant for their cohorts, subjects, stakes, and risk profile. The source set does not settle whether generic AI frameworks are sufficient or whether sector-specific validation should always be required.

- Buying AI for promised efficiency without enough evidence of assessment value. - Locking into opaque systems with hidden labour, data, or environmental costs. - Weak accountability when errors or bias emerge after deployment. - Treating ethical language as a substitute for validation. - Underestimating dependency on vendor platforms, cloud services, or proprietary models. - Adopting systems that fit the procurement narrative but not the assessment purpose.

A sensible approach is to treat procurement as a live hypothesis test. 1. Define the exact assessment problem the AI is meant to solve. 2. Specify what evidence would count as success in the intended context, not just in a demo. 3. Test the hidden costs in labour, data, governance, energy, and dependency. 4. Decide who is accountable if the system is wrong, biased, unavailable, or changed by the supplier. 5. Plan for withdrawal, replacement, or non-renewal if performance falls short. 6. Prepare a justification that can be explained to learners, regulators, auditors, and internal decision-makers. The stronger the stakes, the stronger the need for independent validation and explicit governance.

| Option | When it fits | Main upside | Main risk | |---|---|---|---| | **Prohibit AI** | High-stakes use, weak evidence, or unclear accountability | Lowest dependency and simplest governance | May miss genuine efficiency gains | | **Permit with controls** | Medium-risk use where benefits are plausible and evidence is partial | Keeps human oversight and limits exposure | Controls can become performative if not tested | | **Integrate AI deeply** | Narrower cases with strong evidence, stable workflows, and clear governance | Potentially highest efficiency and consistency | Highest lock-in, change-management, and validation burden | The practical choice is usually not whether AI exists in the workflow, but whether it changes assessment meaning, risk, or accountability enough to require stronger controls.

A qualification team is offered an AI tool to draft feedback comments for assessors. The tool looks efficient in the demo, but the team first checks whether it changes judgement quality, whether it introduces inconsistent tone or bias, and whether assessors can still justify decisions independently. If those questions cannot be answered, the safer choice is a limited pilot with clear success criteria rather than a full rollout.

- Australian Framework for Generative Artificial Intelligence (AI) in Schools. - Helen Beetham, *Automatic for the people?*. - Kate Crawford, *Atlas of AI*. - U.S. Department of Education education leaders AI toolkit.

Vendor and consultancy material often presents responsible AI procurement as a matter of policy alignment, ethical use, and deployment support. That is useful as a market signal, but it does not settle the evidence question. The practical difference for buyers is whether the supplier can demonstrate performance, cost, and governance fit in the real assessment context, not just describe a responsible vision.

### What is AI procurement and governance in assessment? It is the process of selecting and overseeing AI systems so that they support assessment without weakening validity, accountability, or public trust. ### Why does public value matter? Because assessment systems serve learners, institutions, regulators, and the public. A tool that saves time but adds hidden cost or weakens trust may not be good value. ### Is a responsible AI framework enough on its own? Usually not. It can help set expectations, but assessment teams still need local evidence, contracts, and clear accountability. ### What is the main unresolved issue? How much independent proof should be required before an AI system is allowed to influence assessment decisions or operations.

Tim Burnett (Admin)

Test Community Network. "AI procurement, governance, and public value." TCN AI & Assessment Wiki. Last reviewed 2026-05-02. https://www.testcommunity.network/wiki/ai-procurement-governance-and-public-value.html

- Australian Framework for Generative Artificial Intelligence (AI) in Schools. - Helen Beetham, *Automatic for the people?*. - Kate Crawford, *Atlas of AI*. - U.S. Department of Education education leaders AI toolkit.