AI literacy in assessment is best understood as a design problem: can learners and staff use AI in ways that still let an assessment prove what it is meant to prove? The strongest evidence suggests that AI use is already normal in higher education, so policies that assume rare or exceptional use will often be too weak or too strict. The practical question is not whether AI exists, but what level of support, disclosure, and independence each assessment is trying to evidence. UNESCO’s competency frameworks and higher-education research both point towards clearer capability statements, clearer task boundaries, and better governance around acceptable use.

AI literacy in assessment means understanding what AI can realistically do, deciding what support is acceptable, and designing or judging assessment so the evidence still supports valid claims about learner performance. The underlying issue is authenticity: if AI support is now ordinary, assessment rules and task design need to reflect that reality rather than treat AI as marginal or exceptional.

Weak AI literacy can push assessment systems in two bad directions at once: rules that are too restrictive to follow in real learning settings, or rules that are too permissive to protect validity. HEPI reports that 95% of UK undergraduates now use AI and 94% use it for assessed work, while only 36% feel encouraged by their institution to use AI. That suggests many institutions are dealing with a lived reality, not a hypothetical one. Workload and tutoring uses also matter. Practitioner commentary repeatedly frames AI as a time-saver and one-to-one tutor, which changes what learners practise and what independence later assessment is expected to show. UNESCO’s student and teacher competency frameworks strengthen the case that AI literacy is becoming a defined educational outcome, not just a passing digital skill. The Harvard undergraduate survey adds a broader signal that student AI use is now deeply embedded in learner behaviour, which makes this less a niche skills issue and more a mainstream curriculum and assessment question.

- **Authenticity**: whether the work evidences the intended learner performance. - **Acceptable AI support**: what AI use is allowed, disclosed, or prohibited in a given task. - **Assessment vulnerability**: where AI could distort the evidence a task is meant to collect. - **AI pedagogy**: the wider teaching and assessment approach that sets expectations before tools are added. - **Competency framework**: a structured description of what learners or teachers should understand, apply, or create with AI at different levels. - **Human oversight**: keeping human judgement central while AI supports educational work. - **Explainability**: whether staff and learners can understand how AI-supported systems reached a result.

The source set points in a fairly consistent direction: AI literacy is becoming inseparable from assessment design. Stronger sources converge on the idea that literacy is not just tool familiarity, but judgement about what AI can do, where it distorts evidence, and how assessment should respond. There is also broad agreement that institutions need clearer boundaries around acceptable AI support, not just broader enthusiasm for digital capability. The most useful policy direction is not “use more AI”, but “define what level of AI support still preserves the meaning of the result”. UNESCO’s frameworks support this by treating AI capability as something that should be specified, progressed, and assessed deliberately. A further convergence is that governance now matters alongside pedagogy. Sources on human oversight and explainability suggest that AI literacy increasingly includes accountability, transparency, and the ability to explain how AI changes evidence and decision-making.

The main open question is not whether AI matters, but what level of AI support should be normalised in different assessment contexts. Some practitioner commentary leans towards AI as a “second mind” or decision support, but that remains a framing rather than a settled assessment standard. A second unresolved issue is evidence strength. There is a lot of commentary about AI literacy, but relatively little independent evidence showing which literacy interventions actually improve assessment validity, consistency, or learner understanding. The strongest policy sources say capability should be explicit; they do not by themselves settle how that should be assessed in a given qualification or course. There is also a live tension between AI competency and assessment integrity. If AI can outperform students in some exam settings, competence with AI and independence from AI are clearly different claims, and assessment teams still need to decide which claim a task is meant to evidence.

- Rules that are too strict to be followed in real learning and assessment settings. - Rules that are too loose to protect validity. - Uneven interpretation by staff and students, especially where guidance is policy-heavy but example-light. - AI use being policed rather than designed for. - Governance gaps where data protection, sustainability, procurement, and assessment policy are not aligned. - Literacy frameworks being treated as awareness tools only, without changing assessment design. - Human oversight becoming rhetorical rather than operational.

1. Define what the learner must do unaided. 2. Map where AI could support without changing the meaning of the evidence. 3. State what level of AI support is acceptable, disclosed, or prohibited. 4. Test whether staff and learners can apply that rule consistently in practice. 5. Redesign the task if the intended evidence is no longer realistic to protect through policy alone.

### Common assessment stances on AI use | Option | What it means | Best fit | Main trade-off | |---|---|---|---| | Prohibit | AI use is not allowed in the task | High-stakes evidence of independent performance | Can be hard to enforce if AI use is already normal | | Permit with disclosure | AI use is allowed if stated clearly | Coursework, drafting, feedback-rich tasks | Needs clear boundaries or disclosure becomes vague | | Integrate | AI is part of the learning and assessment design | Tasks where AI competence is itself a legitimate outcome | Must separate AI capability from unaided ability | The evidence suggests many institutions are moving away from blanket assumptions and towards explicit task-level decisions. That direction is more defensible than relying on general policy statements alone.

A programme team notices that students are already using AI to draft coursework, but the assignment brief only says “work must be your own”. Instead of tightening the rule alone, the team rewrites the task so students must submit a short process note explaining where AI was used, what was changed, and which parts were done independently. That does not solve every integrity issue, but it gives assessors a clearer basis for judging authenticity and support boundaries.

- HEPI Student Generative AI Survey 2026: strong signal on student behaviour and institutional mismatch. - HE Educator AI Diagnostic: useful practitioner framing of literacy, vulnerability, and design. - UNESCO AI Competency Framework for Students: stronger policy and curriculum signal on explicit AI competencies. - UNESCO AI Competency Framework for Teachers: stronger policy signal on staff capability, ethics, and human-centred AI use. - Anthropic Education Report on how university students use Claude: useful higher-education signal about student AI companionship and task patterns. - LSE Public Policy Review article on proactive approaches to generative AI tools in higher education: useful policy signal on redesign and curriculum planning. - Harvard Undergraduate Survey on Generative AI: useful signal that student AI use is now highly normalised. - Stanford Digital Education keynote on humanity in the loop: useful governance signal about human oversight. - EDEH workshop on explainable AI in education: useful policy and governance signal about transparency and accountability. - BBC reporting on AI outperforming university students in exams: a sharp reminder that literacy and integrity concerns can collide.

The visible market signal in this area is mostly practitioner-led rather than vendor-led. Where vendor material does appear, it usually frames AI as productivity support, tutoring support, or decision support, which is useful context but not validation that a given literacy approach improves assessment outcomes.

### What is AI literacy in assessment? It is the capability to understand what AI can do, define acceptable use, and design or judge assessment so the resulting evidence still supports valid claims about learner performance. ### Why does AI literacy matter in exams or coursework? Because poor AI literacy can lead to either unrealistic bans or weakly protected assessment validity. The issue is not simply tool knowledge; it is whether the task still measures the intended learning. ### Should AI literacy be taught as a separate capability? The source set suggests it may need explicit attention, but there is no settled answer on whether it should sit as a standalone skill or be embedded in wider digital and assessment practice. ### Can AI competency frameworks change assessment practice? Yes, if they are used to define what learners and teachers should understand, apply, and create with AI. But they do not by themselves answer how much AI support should be permitted inside a specific assessment.

Tim Burnett (Admin)

Test Community Network. "AI literacy and assessment design." TCN AI & Assessment Wiki. Last reviewed 2026-05-02. https://www.testcommunity.network/wiki/ai-literacy-and-assessment-design.html

- HE educator AI diagnostic. - HEPI Student Generative AI Survey 2026. - UNESCO AI Competency Framework for Students. - UNESCO AI Competency Framework for Teachers. - BBC News article on AI-generated exam answers outperforming university students. - Anthropic Education Report on how university students use Claude. - LSE Public Policy Review article on generative AI in higher education. - Stanford Digital Education keynote on humanity in the loop. - EDEH workshop on explainable AI in education. - eSchool News on digital testing, AI, and personalised learning. - Test Community Network conversation with Med Kharbach on pedagogy-first AI use. - The Edtech Podcast episode on governance, ethics, and sustainability. - Mr Barton Maths podcast episode with Neil Almond. - Mr Barton Maths podcast episode with James Radburn. - Y-AI? episode with Darren Coxon. - Y-AI? episode with Jane Basnett. - Harvard Undergraduate Survey on Generative AI.