The EU AI Act: When AI Decisions Matter In FE

Using AI to help draft a lesson plan is one thing. Using it to assess whether a prospective learner should study at Level 2 or Level 3 is something else entirely. That difference gets to an increasingly important question for further education. As AI becomes more capable, how much weight should we give its judgements?

EU AI Act and the UK

The European Union’s AI Act offers one way of thinking about that. On 2 August 2026, the Act became generally applicable across EU member states. As the UK is not an EU member state, it does not govern a UK college using AI solely within the UK, although its scope can extend beyond the EU in some circumstances, including where an AI system’s output is used within the Union.

Even so, the European approach is worth watching because it draws a line between everyday uses of AI and those that can seriously affect somebody’s education, opportunities or livelihood. Similar questions are already surfacing in the UK.

Where the level of risk changes

The organising principle of the EU Act is risk. At the lower end are relatively routine uses: an AI tool that organises application files or translates documents would not ordinarily be treated as high-risk. At the other extreme, the Act prohibits practices considered fundamentally unacceptable, including, from 2 December 2026, the use of AI systems to create non-consensual sexually explicit material. Between those ends of the spectrum are applications classed as ‘high-risk’, where stricter requirements around risk management, documentation and human oversight apply.

Education and vocational training feature explicitly. The legislation says AI systems used for certain educational purposes ‘should be classified as high-risk AI systems’. Those purposes include deciding access or admission, evaluating learning outcomes, assessing the appropriate level of education somebody should receive and monitoring prohibited behaviour during tests. The detailed rules are due to take effect from 2 December 2027.

The reasoning is particularly relevant to FE. The Act says systems of this kind can influence the ‘educational and professional course’ of a person’s life and, ultimately, their ability to secure a livelihood. Colleges make such decisions every day. Admissions, programme levels, assessment, progression and qualification outcomes can all shape what a learner does next and which apprenticeship or employment opportunities are open to them.

A key topic of inquiry then is the role that an AI’s output is allowed to play. There is an obvious difference between using AI to generate practice questions and allowing it to contribute directly to a final grade. Likewise, a system that flags where a learner may need additional support is doing something quite different from one whose recommendation helps determine the level at which they are placed. The closer AI gets to decisions with lasting consequences, the harder the questions become: how reliable is the system, who is accountable for its output, and where does human judgement sit?

FE is already using AI in consequential areas

The UK Government’s 2025/26 Cyber Security Breaches Survey found that 82% of the 33 FE colleges surveyed had adopted some form of AI. The sample is small, so the figure needs treating with caution, but it gives some indication of the pace of uptake. Jisc’s 2025 leadership survey gives a clearer sense of where AI is being used. Among FE respondents, 76% reported using it for teaching and learning preparation, 57% for delivery and 37% for assessment, up from 28% for assessment the previous year.

That last figure is particularly significant. AI can support assessment in relatively low-stakes ways, such as creating practice material, suggesting feedback or identifying where a learner may need further support. But, once an AI-generated judgement begins contributing to a formal grade or progression decision, the stakes change. That is broadly the distinction the EU framework is trying to capture.

UK guidance is already asking similar questions

UK colleges do not need to adopt the European framework to take these issues seriously. Much of the groundwork is already being done domestically.

Jisc and the Association of Colleges have developed principles for the use of AI in FE covering safety, transparency and explainability, fairness, accountability and contestability. The last of these is especially important where AI affects an outcome: learners and staff should have a clear way of challenging an output they believe has unfairly disadvantaged them.

The Department for Education’s generative AI product safety standards ask providers and suppliers to consider purpose, evidence, privacy, safeguarding, security, testing and governance.

Ofqual’s July 2026 policy takes a similar approach to qualifications, emphasising valid, reliable and fair outcomes, expert human involvement and accountability. AI cannot currently be used as the sole marker for assessments that form part of regulated qualifications.

What colleges should be asking suppliers

This matters particularly when colleges purchase technology. AI may arrive as automated feedback added to a learning platform, AI-assisted marking within an assessment system or personalised recommendations built into student-support software. Colleges need to know how far its output influences what happens next. Is it providing information to a member of staff, or is its recommendation likely to determine the decision? Can staff understand and override what it produces? Has the supplier tested how it performs across different groups of learners? What evidence supports its claims? And, if a learner believes an AI-supported decision is wrong, what happens then?

None of these questions requires a college to adopt European regulation. The EU Act is useful because it shows why scrutiny should increase as the consequences of a system’s output become more serious.

The UK may still take a different route

There is, at time of writing, no direct UK equivalent of the EU AI Act. The government’s most recently articulated approach has favoured regulating AI at the point of use through existing regulators and laws covering areas such as data protection, competition, product safety and equality.

The debate is far from settled. Parliament’s Joint Committee on Human Rights has recommended a new risk-based AI Bill and said the EU AI Act could provide a ‘useful point of departure’ when considering obligations for higher-risk systems, while stressing that any UK measures would need to be designed for the domestic context.

That is a committee recommendation rather than government policy, so there is no basis for assuming the UK will reproduce the European model. It does show, however, that the approach being taken across the Channel is already part of the discussion here.

For FE, the exact regulatory model may matter less than the underlying question. Further education routinely makes and supports decisions about access, level of study, assessment, progression and what opportunities come next. AI will increasingly sit somewhere within those processes. When colleges consider where and how to use it, a useful starting point is to ask how much this system matters to the person affected by its output.

If it is helping an educator prepare materials or organise information, the answer may be relatively little. If it can influence somebody’s admission, assessment, progression, qualification or future opportunities, considerably more is at stake. In those latter circumstances, colleges will need good reasons for trusting the system, a clear understanding of the evidence behind it, meaningful human judgement and a way for learners to challenge decisions when things go wrong.

By Gary F. Fisher, Academic Developer in Online Education; Liverpool School of Tropical Medicine