For some time, there has been a growing concern that many vocational schools lack a developed, cohesive policy around Artificial Intelligence. While comprehensive survey data specific to the vocational education and training (VET) sector remains scarce, a recent report from the Higher Education Policy Institute (HEPI) sheds light on how UK universities are navigating this complex landscape. The findings are both revealing and, in some ways, uncomfortable. They offer critical lessons for teachers, trainers, and managers in vocational education as they consider how to integrate AI thoughtfully and effectively into their own institutions — and they resonate closely with work already underway in the European VET sector, including through the AI Ready project, in which Pontydysgu is a partner.

The HEPI report, authored by Sam Illingworth, analysed the publicly accessible AI policies of 96 UK higher education institutions. The study aimed to uncover what these policies actually do in practice, contrasting the language they use with the structural functions they serve. The central finding is stark: most university AI policies employ the vocabulary of education and support while operating fundamentally as compliance and surveillance instruments. They promise to foster critical thinking but often deliver audit trails, revealing a significant gap between institutional claims and operational reality [1].

One of the most striking mechanisms identified in the report is what Illingworth calls "performative education framing." Many policies adopt aspirational language about learning and digital literacy, yet their core structures are built around binary lists of acceptable and unacceptable uses, anchored by threats of disciplinary action. A policy might open with a commitment to supporting students in learning how to use AI ethically, only to follow up with a detailed matrix of penalties for academic misconduct. This approach treats AI primarily as a threat to be contained rather than a tool to be understood.

Furthermore, the report highlights the prevalence of "conditional trust." Institutions often require students to declare their AI use, retain evidence of their drafting processes, and submit to verification if requested. In this framework, non-declaration is frequently interpreted as concealment. This adversarial dynamic undermines the development of genuine critical AI literacy. Research consistently demonstrates that trust-based approaches yield better educational outcomes than surveillance-based ones, fostering an environment where students feel safe to engage honestly with new technologies [1].

Another critical insight concerns the importance of a policy's structural location. Where a policy sits within an institution's architecture strongly predicts its framing. Policies embedded within academic misconduct frameworks tend to inherit punitive assumptions, regardless of the educational vocabulary they might employ. Conversely, policies housed within teaching, learning, or study skills sections are more likely to be genuinely educational. The decision of where to locate AI guidance is therefore more consequential than the specific words chosen to articulate it [1].

So, what can vocational schools learn from these findings? The AI Ready project, an Erasmus+ Cooperation Partnership running from October 2025 to September 2028, offers a useful point of comparison. Coordinated by UC Limburg in Belgium and involving partners across Belgium, Greece, and Spain - including Pontydysgu - the project is explicitly built around the principles that the HEPI report identifies as missing from so many university policies [2]. Its mission is to enhance AI literacy by empowering schools, teachers, and students with knowledge about AI's benefits, risks, and ethical use, balancing transformative potential with human-centric values. Crucially, the project frames AI policy not as a compliance instrument but as a framework for responsible and informed practice.

One of the AI Ready project's central outputs is the development of comprehensive guidelines to help schools implement responsible AI policies. This mirrors the HEPI report's call for institutions to move beyond performative framing and produce guidance that genuinely educates. The project also aims to enhance the European Commission's SELFIE tool with thirty new AI readiness assessment questions, providing schools with a structured means of understanding where they stand and what they need to develop. This kind of sector-wide coordination is precisely what the HEPI report finds lacking in UK higher education, where 163 institutions appear to be developing AI policies in isolation, producing wide variation in quality and orientation [1].

The AI Ready project's approach to teacher development is equally instructive. Rather than issuing top-down compliance guidance, it offers a continuing professional development course covering AI fundamentals, ethics, teaching strategies, and practical applications, complemented by transnational workshops in which educators from three countries share practice. This collaborative, cross-border approach stands in contrast to the fragmented, institution-by-institution picture the HEPI report describes. Students, too, are treated as active participants rather than subjects of policy: the project organises interactive workshops exploring AI applications and ethics, and provides a student handbook built around the KWL learning method — what students know, what they want to know, and what they have learned.

These parallels point to a broader lesson for vocational schools. First, AI policy should be integrated into teaching and learning frameworks rather than appended to misconduct procedures. In VET, where the focus is on practical application and industry readiness, students need to understand not just how to use AI tools but how to interrogate their outputs, biases, and implications in real-world professional contexts. Secondly, trust must be the default position. Surveillance and automated AI detection are not only technically unreliable but also operationally expensive and structurally unfair. Rather than investing in an arms race of detection and evasion, institutions should focus on assessment design. If an assessment can be easily completed by a chatbot, it is the assessment that needs redesigning. Thirdly, policy development should involve students from the outset. A policy written without input from those it affects will reproduce institutional assumptions by default.

Finally, and perhaps most importantly, vocational schools should prioritise critical literacy over mere tool proficiency. Teaching students how to prompt a specific AI model is compliance training. Educating them to question the assumptions, environmental impacts, and ethical considerations embedded in AI systems is education. The AI Ready project embodies this distinction, and the HEPI report confirms that too few institutions in higher education have managed to make it in practice. As vocational schools develop their own approaches - whether independently or through collaborative projects such as AI Ready — they have the opportunity to learn from the higher education sector's missteps and build something more genuinely fit for purpose.

References

[1] Illingworth, S. (2026, May). What UK university AI policies actually do: A study of 96 institutions (Policy Note 71). Higher Education Policy Institute. https://www.hepi.ac.uk/wp-content/uploads/2026/05/What-UK-university-AI-policies-actually-do-A-study-of-96-institutions.pdf

[2] UC Limburg. (2025 ). AI Ready: Empower schools, teachers and students to be AI ready [Erasmus+ Cooperation Partnership project website]

About the Image

Museums are using AI to speed up the publishing of contextual descriptions for artefacts. Yet, in order to describe visual objects in museums, we must do more than simply identify and name objects, but also acknowledge their cultural and historical context. The ability for AI models to generate such information ultimately depends on the quality of the underlying datasets. Furthermore, datasets are usually Western European-centric, meaning that using AI to describe images of Hindu deities or African museum artefacts may be less effective and also be described from a European perspective. The image raises awareness of the importance of developing better knowledge bases of cultural heritage and museum collections, before working on AI systems to describe collections.