Policy Brief: Regulating Student Generative-AI Use in Assessment

Demonstration writing sample. This brief illustrates my policy-writing style. Recommendations are drafted for demonstration; verify all claims against current evidence before institutional use.

The issue

Generative AI can now produce credible drafts of most text-based assessment tasks. Institutions must decide what student AI use is legitimate, how it must be declared, and how breaches are established — under conditions where reliable technical detection does not exist.

Why detection-led responses fail

Detection tools produce false positives and false negatives at rates that cannot support high-stakes decisions.
Detection frames the problem as an arms race, deferring the real questions of what students should learn and demonstrate.
Enforcement built on unreliable evidence creates procedural-fairness risk for institutions and genuine harm for wrongly accused students.

What a workable policy architecture contains

Task-level permission statements. Rules set where students meet them — in the assessment task — within an institutional framework, not instead of one.
A disclosure mechanism that is safe to use. Declarations must be simple, and honest disclosure of permitted use must never be treated as an admission.
Authorship-based evidence practice. Where substitution is suspected, establish whether the student can account for the work — through drafting evidence and structured conversation — rather than relying on detector scores.
Proportionate outcome categories. Distinguish acceptable use, disclosure failures, and substitution of the student’s work; respond to each differently.
Assessment redesign as the long-term control. Tasks that require demonstrated process, judgement and defence of one’s work reduce the enforcement burden policy alone cannot carry.

Recommendation

For a leadership team acting on this brief, the priority order is: publish a single task-level permission statement template that units can adopt without drafting their own from scratch; pair it with a disclosure mechanism piloted with student and staff feedback before mandating it sector-wide; and shift detection-focused investment into assessment redesign, which reduces the enforcement burden a policy alone cannot carry. Sequencing matters more than completeness — a narrow, working rule that closes the most common ambiguity first is more useful than a comprehensive policy that arrives a year late.