In March 2026, RAND published data that should be pinned to every staffroom noticeboard: 62% of students now use AI for homework – up from 48% just seven months earlier.

That’s not a trend.

That’s a transformation, happening faster than any curriculum cycle, policy review, or professional development calendar was built to handle.
Here’s what makes that number complicated. Ask those same students whether AI is hurting their thinking and 67% will say yes – up more than 10 points in less than a year. Your students are not oblivious. They know something is being traded away. They just don’t know what to do about it, because more than 80% of them say no teacher has ever actually taught them how to use AI well.

The system gave them the tool. Nobody gave them the lesson.

Meanwhile, assessment hasn’t moved. Most tasks still reward the product – the essay, the worked solution, the submitted report – rather than the process that produced it. In a world where any student with a phone can generate a plausible product in 30 seconds, grading the product alone tells you almost nothing. It doesn’t tell you whether they understood it, questioned it, or learned anything at all.

Research published in early 2026 is converging on the same design principles: move toward process-based, performance-based, and reflective assessment. Not because they’re harder to cheat — though they are — but because they actually measure what we claim to care about: reasoning, judgement, the ability to explain your own thinking under pressure.
This matters beyond academic integrity. Jenny Anderson and Rebecca Winthrop’s research across 65,000 students found that fewer than 1 in 10 students ever operate in genuine Explorer mode – curious, self-directed, intrinsically motivated. The rest are passengers. AI, used without intentional pedagogy, optimises the passenger experience: faster compliance, zero growth.

Assessment is the lever. When you change what you ask students to demonstrate, you change what they practice. When you change what they practice, you change what they become capable of.

The data isn’t waiting for the next policy cycle.

Neither should you.