Detection-led responses to generative AI are unsustainable at institutional scale.
The Institutional Challenge
Generative AI has fundamentally altered the landscape of student assessment. Tools capable of producing high-quality written work, code, analysis and creative output are freely available and improving at pace.
Most institutions have responded with detection-first strategies: AI detection software, revised submission policies, and increased invigilation. These measures address symptoms without confronting the underlying structural problem.
The challenge is not technological; it is architectural. Assessment models designed before generative AI assumed conditions that no longer hold.
Why Detection Fails as a Strategy
Detection-based approaches create compounding institutional problems:
False positives undermine trust between students and institutions
Detection accuracy degrades as AI outputs become more sophisticated
Staff time is consumed by procedural escalation rather than academic improvement
Inconsistent application across schools creates governance risk
Over-reliance on detection tools shifts institutional responsibility to third-party vendors
Detection may retain a supplementary function, but it cannot form the foundation of an institutional integrity strategy.
A Design-Led Response
Sustainable assessment reform requires a shift from monitoring outputs to designing assessment that is inherently resistant to AI substitution or outsourcing. This involves:
Embedding process-oriented assessment that captures iterative student work
Introducing authentic evaluation tasks tied to context-specific application
Aligning assessment briefs with programme learning outcomes rather than generic competency statements
Building staff confidence and capability in assessment redesign
Establishing clear institutional frameworks for permitted and prohibited AI use within assessment
Redesigned assessments do not eliminate AI from the process. They render it visible and accountable, shifting the emphasis from policing to learning design.
Governance Alignment
Assessment redesign cannot operate independently of institutional governance. Changes to assessment frameworks, marking criteria and submission processes should be aligned with quality assurance frameworks, programme approval mechanisms and regulatory expectations.
Without governance alignment, redesigned assessments risk creating inconsistency across schools and programmes, replacing one set of problems with another.
Conclusion
The question facing universities is not whether to adopt AI detection, but whether their assessment architecture remains defensible in a post-AI context. Institutions that invest in design-led reform, sequenced through governance and supported by staff capability, will build systems that protect standards without reliance on increasingly unreliable detection tools.