COMPLIANCE · CHECKLIST
JCQ and Ofqual AI Guidance: A Practical Checklist for Private Training Providers
Reading the guidance and acting on it are two different skills. Something you can walk through on a Tuesday afternoon and come out knowing whether you’re covered.
By Eduface · July 2026 · 9 min read
Reading regulatory guidance and actually acting on it are two different skills, and most training providers we talk to are pretty good at the first one and stuck on the second. You’ve read the Ofqual working paper. You’ve seen the JCQ guidance referenced in three different vendor pitches this quarter. What you actually want is something you can walk through on a Tuesday afternoon and come out the other side knowing whether you’re covered. That’s what this is.
Start with what you’re actually being asked to prove
Strip away the language and Ofqual and JCQ are asking training providers to demonstrate two separate things, and it helps to keep them separate in your head rather than treating AI compliance as one blob.
The first is about your assessors: is a human genuinely making the marking decision, or is AI making it for them with a human rubber-stamping the result afterwards. Our companion pieces on Ofqual’s marking rules and on human-in-the-loop marking cover this in depth.
The second is about your learners: are you confident the work submitted is theirs, given how easy it now is to generate a plausible-sounding assignment with the same tools your assessors might be using to mark it. JCQ guidance puts the responsibility for this squarely on the centre, not the awarding organisation and not the student in isolation.
Treat these as two separate audits, because a tool that solves one doesn’t automatically solve the other.
The checklist, assessor side
Can you point to a specific, mandatory step where a human assessor sees the AI’s reasoning and explicitly approves a grade, with no path for it to reach a learner without that step happening? If you can’t answer this in one sentence, that’s worth flagging as a gap rather than assuming it’s fine.
Do you have a record, for each submission, of what the AI proposed and what the assessor actually decided? Not just that the software ran. Evidence that a specific person engaged with that specific piece of work.
Has anyone on your team actually asked your awarding organisation, in writing, what their current position is? Ofqual’s framework explicitly leaves room for awarding organisations to set their own rules, and expects real variation between them, so a policy that applies to one qualification you deliver doesn’t necessarily apply to another.
The checklist, learner side
Do your learners know what counts as acceptable AI use and what doesn’t, in terms specific enough that a first-year adult learner returning to study after a decade away could actually follow them? JCQ guidance requires centres to ensure understanding, not just publish a policy nobody reads.
If a submission looks like it may not be a learner’s own unaided work, is there an actual process for raising that, or does it rely on an assessor happening to notice and deciding, on their own initiative, to make an issue of it?
Is your approach to this consistent across assessors, or does it depend heavily on which individual happens to be marking that batch of work? Consistency here isn’t a nice-to-have. It’s the difference between a defensible policy and an arbitrary one if a decision is ever challenged.
If you also work in Ireland or elsewhere in the EU
Most of what’s above is UK specific, built around Ofqual and JCQ. If you also deliver qualifications into Ireland or elsewhere in the EU, there’s a separate and genuinely more demanding framework worth knowing about. The EU AI Act (Regulation 2024/1689) classifies AI systems used to evaluate learning outcomes as high risk under Annex III, and high risk systems come with formal obligations around risk management, data governance, transparency, and documented human oversight. That classification isn’t a distant future concern either, the Act’s provisions for these systems became fully applicable in August 2026.
One detail worth sitting with: guidance on the Act has been explicit that a human review step only keeps a system out of the high-risk bracket if that review is genuinely independent, not a rubber stamp on an AI decision that’s already effectively been made. That’s the same substantive test Ofqual applies, arrived at through a different legal route. If your workflow satisfies one, it’s most of the way to satisfying the other, but it’s worth checking both rather than assuming.
What good documentation actually looks like
A lot of providers treat compliance as something you demonstrate reactively, if someone ever asks. That’s backwards, and it tends to fall apart exactly when it matters most, during an appeal or an awarding organisation audit, when memory and good intentions aren’t enough. Keep a simple, dated record of the questions you asked your awarding organisation and the answers you got. Keep evidence, even lightweight evidence, that assessors are engaging with AI-drafted grades rather than approving on autopilot. And revisit this checklist roughly quarterly, because both Ofqual and the EU AI Act have said explicitly that guidance in this area is going to keep developing.
Frequently asked questions
Is this checklist enough on its own to guarantee compliance?
No single checklist can guarantee that, since your specific awarding organisation may have additional requirements. Treat this as a starting point for a conversation with them, not a substitute for one.
Do I need to worry about the EU AI Act if I only deliver qualifications in the UK?
Not directly, since it’s an EU regulation. It becomes relevant if you deliver into Ireland or elsewhere in the EU, or if your awarding organisation operates across both jurisdictions.
What’s the single biggest gap you see when providers audit themselves against this?
Evidence. Providers are often doing the right thing in practice, an assessor really is reviewing each grade, but they have no record of it happening, which leaves them exposed if a decision is ever challenged.
How often should we revisit this?
Roughly quarterly. Both Ofqual and the EU AI Act’s implementing guidance are explicitly still developing, and a policy read once in early 2026 won’t stay accurate indefinitely.
Sources
Ofqual. Ofqual’s Approach to Regulating the Use of Artificial Intelligence in the Qualifications Sector. gov.uk.
JCQ. AI Use in Assessments: Protecting the Integrity of Qualifications.
Regulation (EU) 2024/1689 (EU AI Act), Annex III, point 3(b) and Article 6.
Close the evidence gap on marking
Eduface pairs every AI draft with the assessor’s decision in an exportable audit trail. Book a demo or start free.