Foresight or blindsided? – Measured Response #1

Welcome to the first edition of Measured Response

The name is deliberate. Assessment is, at its core, an act of measurement, of knowledge, of competence, of readiness. But the sector right now needs more than measurement; it needs measured thinking. AI is rewriting what’s possible and what’s at risk. Cheating is being industrialised. Item banks are being harvested. And underneath all of it sits an older, harder question: are/were our assessments ever as robust as we believed?

This is a weekly blog about assessment, AI, test security, and whatever else in the sector catches my eye. Some weeks it’ll be a considered take on a long-running debate. Other weeks it’ll be a sharper response to something that’s just caught my eye in the sector. Either way, the aim is the same: to think clearly, in public, about a sector I’ve spent more than twenty-five years inside.

I’m Paul Muir, Chief Customer Officer at risr/, Board Chair of the Association of Test Publishers (ATP), Ambassador for The Cambridge Assessment Network and official ice-cream taster for the United Kingdom.

The views here are my own; measured, but not always neutral.

Week 1 – Foresight or blindsided?

In this first edition, I want to rewind 48 hours and reflect on a fantastic session with the Certification Network Group (CNG) in Chicago. A session that, in its preparation, used environment scanning and foresight methods across 10 different AI models to ask “How should AI be applied in the next 12 months?” in our industry.

So, what does AI think about how AI will be applied within our sector across the next 12 months? And importantly, Isn’t that a biased question? And do we agree?

The top five areas according to the research:

1. AI-assisted exam content
2. AI-driven integrity assurance
3. Personalised candidate preparation
4. Automated scoring
5. AI-powered Job Task Analysis & Blueprinting

So, is the list right?

My first reaction, sitting in the room in Chicago: of course it is. These are the conversations our industry is having at every conference, in every client meeting, on every steering group agenda. The models didn’t surprise us, they simply reflected us.

Which is exactly the problem.

Ten AI models were asked how AI will be applied in our sector. We got back a tidy summary of what our sector is already doing with AI. That’s not foresight. That’s a mirror.

With that caveat parked, let me take the five in turn:

1. AI-assisted exam content. Right answer, slightly flattering reason. It’s top of the list because it’s the most adopted, not the most transformative. The transformation comes when provenance and SME accountability become baseline expectations rather than vendor differentiators. Think about “Expert in the Loop”, not just a “Human in the Loop”.
2. AI-driven integrity assurance. This is where I think the list under-reads the moment. I’ve spent the last few years debating test security and the impact of AI and too often Integrity assurance is framed as a defensive action: anomaly detection, pattern recognition, catching the cheats. Fine. But the integrity conversation in 2026 is no longer about catching, it’s about designing for a world where catching is structurally insufficient.
3. Personalised candidate preparation. I’d push back the hardest here. Personalised prep is a learning problem, not an assessment problem. More importantly, personalised prep powered by AI is where the equity gap widens fastest: candidates with better tools, better prompts and better access get better prepared. That’s not a feature of the future I want to see, it’s a fairness problem we haven’t named yet. I’d drop this out of the top five and replace it.
4. Automated scoring. Possibly under-ranked. Performance-based assessment was logistically prohibitive for a generation; automated scoring is what I believe makes it viable at scale. If our sector accepts that recall-heavy assessments were never as robust as we claimed, then automated scoring isn’t a productivity play, it’s the thing that lets us assess what we should have been assessing all along. I’d move this up the list.
5. AI-powered JTA & Blueprinting. This is the quietly radical one and the list buries it at number 5. JTA and blueprinting are where construct validity is defined. Getting that layer right is what makes everything above it defensible. If we automate item generation on top of ‘stale’ blueprints, we’re scaling the wrong thing faster. I’d move this up too, possibly to the top.

What I’d choose instead

If I were writing the list rather than reading it, I’d reorder, and I’d swap one out. Here’s my Measured Response:

1. AI-powered Job Task Analysis & Blueprinting. Get the construct right, or nothing above and beyond it matters.
2. Assessment redesign (not just integrity assurance). Stop catching. Start designing authentic assessments AI can’t trivially solve.
3. AI-assisted exam content (with mandated provenance). Scale, yes. Unaccountable scale, no.
4. Automated scoring for performance-based assessment. The unlock for authentic assessment and measurement.
5. Regulatory and equity readiness. The one the list missed. EU AI Act, high-risk classification, candidate access disparities, audit-ready vendor documentation. None of this is glamorous. All of it lands in the next 12 months whether we plan for it or not.

Personalised candidate preparation comes out. Not because it doesn’t matter, because it isn’t an assessment-sector priority. It’s an ed-tech priority we keep adopting by mistake.

The deeper point: the AI didn’t get the list wrong. It gave us back a faithful summary of what our sector is already doing. The job of foresight isn’t to confirm where we are. It’s to ask what we’re not looking at.

That’s the Measured Response.

Over to you…

I want Measured Response to be a forum for debate and discussion. It’s about opinions, it’s about thinking about what we would do differently as we all have different perspectives and experiences.

Tell me which of the original five is wrong. Which of my five is wrong, and what would you swap in?

Comments are open.

Until next week,
Paul Muir