Ofqual keeps existing AI rules for England’s qualifications sector

Ofqual on Thursday updated its policy paper on artificial intelligence, saying it will keep regulating AI in England’s qualifications sector through its existing outcomes-based framework rather than by writing a new AI-specific rulebook. The practical message to awarding organisations is permissive but strict: they may use AI, but they must be able to show that it produces valid, reliable and fair outcomes, with “appropriate expert human involvement and oversight, quality assurance and accountability for outcomes.” (gov.uk)

The update matters because it is the clearest current statement of how England’s exams regulator wants the sector to handle AI as use spreads from administration and assessment design into more sensitive territory such as marking, moderation support and malpractice detection. Ofqual says this regulatory approach has been in place since April 2024 and remains grounded in five objectives: fairness for students, validity of qualifications, assessment security, public confidence and innovation. (gov.uk)

The policy applies to England and is aimed first at awarding organisations, not directly at schools or colleges. But school and college leaders will feel the effects anyway, because exam board decisions about assessment design, marking processes, non-exam assessment controls and acceptable use of AI ultimately shape what teachers can set, supervise and authenticate. Ofqual’s own guide for schools and colleges says the regulator does not place requirements directly on schools, even as it regulates the qualifications and assessments they deliver. (gov.uk)

No new AI rulebook, but not a light-touch regime

What changed on July 16 is better understood as a clarification and refresh of Ofqual’s stance than as a sudden new legal regime. The updated paper explicitly says the regulator will apply its existing framework to AI rather than introduce AI-specific rules, and it repeats that the approach has been in force since April 2024. In other words, awarding organisations should not read the paper as a green light to move faster; they should read it as notice that old obligations still apply when the tool is new. (gov.uk)

That position is easier for Ofqual to take because its broader rulebook already reaches the questions AI raises. The Ofqual Handbook spans governance, risk, qualification design, assessment delivery and marking, while Section H requires effective arrangements so marking criteria are understood and applied consistently. Since December 2025, the framework has also included a Principles Condition requiring awarding organisations to treat learners fairly, ensure qualifications are fit for purpose, maintain public confidence and take a proactive approach to compliance.

For boards, that means the compliance burden is likely to fall less on whether a system is marketed as “AI” and more on whether they can evidence what it does, where it can fail, who checks it and how decisions can be challenged. That is an inference from Ofqual’s paper, but it is a strong one: the regulator is asking for outcomes, oversight and accountability, not vendor promises. For procurement teams and assessment leaders, that points toward more documentation, more validation work and tighter internal sign-off before AI tools touch any high-stakes process. (gov.uk)

The clearest boundary remains marking. In January, Ofqual published a 104-page working paper saying the use of AI as the sole mechanism for determining a student’s mark does not comply with current regulations. At the same time, the regulator said there are “promising applications” for AI in quality assurance of human marking and in training new markers, provided human expertise and judgment remain central. (gov.uk)

That caution reflects more than institutional conservatism. Ofqual’s research says current large language models lack true semantic understanding and human-like judgment, can behave unpredictably and often operate as “black boxes,” making it hard even for experts to explain specific outputs. In a sector where grades affect progression to sixth form, university, apprenticeships and jobs, explainability is not a nice-to-have. It is part of the case for why a result should be trusted. (gov.uk)

The student cheating problem is separate, but connected

The other half of the story is student misuse of AI, which Ofqual is treating as a different regulatory problem from boards’ own use of the technology. In March, Chief Regulator Sir Ian Bauckham wrote to exam board chief executives asking for stronger measures on malpractice, highlighting misuse of AI in non-exam assessment alongside mobile phones in exam halls. In April, Ofqual followed with an AI malpractice advice note stressing that it does not create new requirements; instead, it explains how existing conditions already apply to AI-related risks. (gov.uk)

That distinction matters for schools. Ofqual’s March resources for schools and colleges say plainly that using AI to produce coursework and passing it off as a student’s own work is cheating, and the regulator has published lesson materials and briefing packs to help leaders talk to students about assessed work. The same package also underscores why the issue is so sensitive: if AI does the work, the grade stops meaning what users of the qualification think it means. (gov.uk)

The likely second-order effect is more pressure on awarding organisations to review which assessment formats are most vulnerable to AI-assisted malpractice, especially where work is produced over time, remotely or with limited authentication. Ofqual’s AI malpractice note says risk will vary by how assessments are designed, delivered and used. That does not amount to a ban on coursework, but it does suggest that future qualification design will put a premium on authenticity controls, supervised elements and clearer rules about acceptable support. (gov.uk)

Innovation is allowed, but trust comes first

Ofqual is not trying to freeze the sector in place. The updated paper says one purpose of the policy is to enable safe AI-driven innovation and growth, and the 2025 Principles Condition was partly sold as a way to help organisations handle new or novel situations without constant rewrites to the rulebook. That combination gives boards room to experiment with lower-risk uses of AI, particularly where the technology supports human decision-making rather than replaces it. (gov.uk)

Still, the trade-off is clear. A flexible, outcomes-based regime can adapt faster than prescriptive rules, but it also leaves awarding organisations carrying more responsibility for proving that a tool is acceptable before they deploy it. The updated paper itself says AI use in this context is still emerging and its implications for assessment are “not yet fully understood.” For the sector, that means uncertainty is not a gap in the policy; it is part of the policy. (gov.uk)

So the immediate takeaway for exam boards is not that Ofqual has opened a new fast lane for AI. It is that the regulator has drawn the lane markings more clearly: innovation is possible, automation is not banned, but any system that cannot be defended on fairness, validity, security and public confidence grounds is unlikely to survive regulatory scrutiny. The next thing worth watching is whether awarding organisations now publish more detailed evidence and operating models for AI-assisted assessment work, especially in marking support and in qualifications that still rely on extended coursework or other vulnerable forms of non-exam assessment. (gov.uk)