Non-exam assessment authenticity is about whether coursework, extended writing, portfolios, and similar tasks can be trusted as the learner’s own work. Generative AI has made this a design issue, not just a misconduct issue, because polished submissions can now be produced with limited visible learner judgement. The central question is how much of the learner’s thinking, drafting, and decision-making is actually evidenced in the assessment. Stronger practice increasingly relies on process evidence, not just the finished product.

Non-exam assessment authenticity is the extent to which a non-exam submission genuinely reflects the learner’s own performance. It matters most where the construct includes independent reasoning, drafting, revision, or authorship, rather than only the quality of a final artefact. AI writing tools make that distinction harder to see, which is why authenticity has become a core assessment design concern.

For assessment leaders, this is a validity and trust issue as much as a misconduct issue. If a task can be completed with substantial AI help and still appear polished, the qualification may be measuring output quality without adequately evidencing learner thinking. That affects how much weight can safely sit on coursework, how tasks are designed, and how moderation and authentication are managed.

- **Authenticity**: whether the submitted work reflects the learner’s own performance, not just a plausible final artefact. - **Process evidence**: drafts, checkpoints, oral defence, referencing, sign-off, or other traces that show how the work was produced. - **Product versus process**: AI may be less problematic where the construct is the finished product, and more problematic where independent reasoning or authorship is the construct. - **AI assistance**: support that may alter how much learner judgement is visible in the final submission.

The stronger evidence points in the same direction: authenticity is increasingly a design problem, not just a surveillance problem. Where coursework contributes meaningfully to a qualification, the assessment needs to show how the learner’s own process is evidenced, not merely how the final artefact is checked. Controls such as checkpoints, staged sign-off, stronger referencing, and footnoting all move towards making thinking more observable. There is also broad practical agreement that polished end products alone are a weaker basis for confidence when generative AI can produce convincing text with little visible reasoning. Practitioner commentary and public demonstrations both reinforce the same direction of travel: assessment design needs to expose learner judgement more explicitly.

The open question is not whether AI changes authenticity, but how far different forms of coursework can still be trusted and what controls are proportionate. The source set does not settle when AI assistance becomes disqualifying, or which combinations of task design and supervision are sufficient for different qualification contexts. The hardest unresolved issue is operational: how much added process evidence improves trust without creating unsustainable workload for learners and assessors. A further open question is how far AI detection can support qualification decisions.

- **Validity risk**: the assessment may overstate independent learner capability if the final product hides heavy AI assistance. - **Fairness risk**: learners with better access to AI tools may gain an advantage if the task does not require visible reasoning. - **Workload risk**: more checkpoints, sign-offs, and source tracing can raise administrative burden. - **Security and trust risk**: if authenticity controls are weak, confidence in coursework can fall. - **Design risk**: removing coursework may protect authenticity but can also remove valuable evidence of learning.

1. Define what the task is meant to evidence: final product quality, independent reasoning, authorship, research, or revision. 2. Decide how much AI assistance is acceptable for that construct, rather than assuming one rule fits all coursework. 3. Build in process evidence where learner judgement matters: checkpoints, drafts, oral defence, controlled drafting, or documented sign-off. 4. Make the permitted use of AI explicit so learners know whether the assessment is AI-assisted, AI-tolerant, or AI-dependent. 5. Review whether the task still gives defensible evidence of learner performance, rather than relying on detection alone.

| Option | What it means | Strengths | Trade-offs | |---|---|---|---| | **Prohibit AI** | Candidates produce work without AI assistance | Clearer authorship expectations | May be hard to enforce and may remove useful tools | | **Permit AI with controls** | AI use allowed within defined rules and disclosure | More realistic and flexible | Requires stronger task design and moderation | | **Integrate AI into the construct** | The assessment expects responsible AI use as part of the competence | Can reflect real-world practice | Needs careful validity framing and clear marking criteria | The key choice is not simply whether AI is allowed, but whether the qualification is trying to evidence unaided performance, supported performance, or responsible use of AI.

A college asks students to submit a reflective report on a project. The final document looks polished, but there is little evidence of drafting or decision-making. The team adds short checkpoints, annotated drafts, and a brief viva so assessors can see how the student reached the final argument. That gives a stronger basis for judging authorship without depending on a detection tool.

- FE Week / Schools Week reporting on AI risks to coursework and Ofqual’s response. - Practitioner commentary on making student judgement visible. - Smart-glasses demonstration article describing AI-assisted exam answering.

The vendor footprint in the source set is limited. The market signal is that tooling and policy are moving faster than settled evidence on what degree of AI assistance can be tolerated in different assessment designs. Any supplier claim that a product can “solve” authenticity should be treated cautiously unless supported by independent evaluation.

**What is non-exam assessment authenticity in assessment?** It is the extent to which coursework or similar work can be trusted as the learner’s own performance rather than AI-assisted output. **Why does authenticity matter for coursework and portfolios?** Because coursework can influence grades and decisions, and weak authenticity design can mean the assessment measures polished output rather than genuine learner judgement. **Can AI be used safely in coursework?** Sometimes, but only where the qualification design makes clear what level of assistance is acceptable and where the learner’s own thinking is still visible. **What are the main risks to watch for?** Hidden AI assistance, weak evidence of authorship, unfair advantages, and increased workload if extra controls are added late.

Tim Burnett (Admin)

Test Community Network. "Non-exam assessment authenticity." TCN AI & Assessment Wiki. Last reviewed 2026-04-22. https://www.testcommunity.network/wiki/non-exam-assessment-authenticity.html

- FE Week / Schools Week reporting on AI risks to coursework and Ofqual’s response. - LinkedIn post by Brooke Ashlee McKinney on making student judgement visible. - Smart-glasses demonstration article describing AI-assisted exam answering.