Durham's AI Assessment Scale Wins New Funding as Schools Confront How to Grade in the Age of Generative AI

A Small Grant With an Outsized Question Behind It

On June 18, researchers at Durham University announced a 25,000 pound grant from UK Research and Innovation's Economic and Social Research Council to expand their work on AI assessment in Irish secondary schools. The headline number is modest, but the question it funds is one that every institution and employer is now wrestling with: when generative AI can produce competent essays, code, and analysis on demand, how do you assess what a person actually knows? The project, led by Durham's Dr Jasper Roe with Dr Mike Perkins of British University Vietnam, will run a live pilot inside Irish schools rather than imposing rules from above, testing whether structured assessment frameworks hold up in real classrooms.

We think the framing matters more than the funding size. Most of the public debate about AI and assessment has been stuck in a binary loop of detect-and-ban versus surrender. The Durham team's premise is that neither extreme works, and that the productive path is designing assessments around declared, calibrated levels of AI use. Dr Roe described the ambition plainly: "We hope this project will help drive impact and support AI readiness in the K-12 landscape, and lead to new frameworks, insights, and practical tools for teachers, students, and educational leaders." That is a procurement-grade goal, not just an academic one.

The AI Assessment Scale Is Quietly Becoming a Standard

The work builds on a tool the researchers developed called the AI Assessment Scale, which gives educators a structured way to specify how much AI use is permitted in a given task, from none to fully integrated. Its traction is the real story: the scale has been translated into more than 30 languages and is now used in over 350 schools and universities internationally. That kind of organic, cross-border adoption is how de facto standards emerge, and it has happened without a vendor, a mandate, or a marketing budget behind it. When a framework spreads that way, it is usually because it solves a problem people were desperate to solve and offered nothing better.

For enterprise learning and development leaders, the parallel is direct. Corporations running internal certifications, compliance training, and skills assessments face exactly the integrity problem that schools do, only with higher financial stakes. A salesperson who passes a product-knowledge exam using a chatbot has not actually internalized the product. The AI Assessment Scale offers a vocabulary for designing assessments that either neutralize AI assistance or explicitly incorporate it as a measured skill. We expect the corporate training market to borrow heavily from this academic groundwork, because the alternative, pretending AI is not in the room, is no longer credible.

Why a Live School Pilot Beats a Policy Memo

The new funding specifically supports moving from theory into the messy reality of working schools. The pilot brings in serious institutional partners: the Joint Managerial Body for Voluntary Secondary Schools, Ireland's Department of Education and Youth, and the State Examinations Commission. The plan is to extend the pilot across participating secondary schools, develop teacher-training programmes, produce practical classroom resources, and expand the evidence base. Notably, the approach examines how frameworks function within live settings rather than dictating institution-wide rules, a recognition that assessment integrity is built in classrooms, not in compliance documents.

Dr Perkins emphasized the shift toward usable tools over abstract argument: "The discussion reinforced the importance of moving beyond abstract debates about AI and assessment, and towards practical frameworks that teachers, students, and school leaders can actually use." This is the same gap that plagues enterprise AI governance, where polished policy documents sit on intranets while employees quietly do whatever the tools allow. The lesson is that frameworks only change behavior when they are translated into the specific, day-to-day decisions that frontline people actually make. A grant that funds teacher training and classroom resources is buying adoption, not just research.

Assessment Integrity Is Now a Procurement Category

The reason CTOs and CIOs should care about a school-assessment project is that assessment integrity has quietly become a cross-cutting procurement concern. Any organization that issues credentials, gates promotions on certifications, or relies on test scores to validate competence now has a generative-AI exposure it did not have three years ago. Education-technology vendors are racing to build proctoring, authentication, and AI-aware assessment features, and buyers need a conceptual framework to evaluate them. The AI Assessment Scale provides one, and the fact that public-sector education bodies are formally backing its validation lends it the kind of institutional credibility that proprietary vendor tools lack.

There is also a strategic read here about where genuine standards in AI education will come from. The flashy announcements come from the large platforms, but the durable frameworks, the ones that shape how millions of assessments are actually designed, are emerging from underfunded academic work with names few executives will recognize. We would watch the AI Assessment Scale more closely than many better-capitalized launches, precisely because its adoption curve suggests it is meeting a real need. For leaders building skills and certification programmes, the move is to align internal assessment design with frameworks that are gaining cross-institutional traction, rather than waiting for a vendor to sell them a black box.

The Bigger Picture for Skills Validation

Step back and the Durham project sits at the center of a problem that will define the next decade of workforce development: the divorce between credentials and competence. For a century, organizations have trusted that a passing grade, a degree, or a certificate is a reliable proxy for capability. Generative AI snapped that link almost overnight, and no institution has fully replaced it. The schools piloting the AI Assessment Scale are, in effect, running the experiments that will tell the rest of us which assessment designs still mean something. Their findings about what works in a classroom of teenagers will travel directly into how companies design the exams that gate hiring, promotion, and regulated professional licensure.

The uncomfortable implication for technology leaders is that some of their existing skills data is already compromised. Certifications earned in the past two years, online assessments completed without supervision, and self-reported competency scores may overstate what employees can actually do unaided, which matters enormously when those same employees are being asked to supervise AI systems rather than be replaced by them. Investing now in assessment designs that distinguish genuine capability from AI-assisted output is not an academic nicety; it is risk management. The 25,000 pound grant is small, but the question it funds, how to know what a person truly knows, is one of the most consequential in the entire skills economy.