Rethinking Assessment for Engineering Students in Higher Education in the Age of Generative AI: A Critical Narrative Review

Abstract

Generative artificial intelligence (GenAI) has unsettled a central premise of higher-education assessment: that the quality of a submitted artefact is a sufficiently trustworthy proxy for the competence of the named student. This problem is acute in engineering, where text, code, calculations, models, design rationales and technical reports can increasingly be generated or transformed by general-purpose and specialised artificial intelligence (AI) systems. This critical narrative review examines how assessment in higher engineering education should be reconceptualised when GenAI is simultaneously a learning resource, an emerging professional tool and a source of construct-irrelevant assistance. Literature published from 1 January 2018 to 27 June 2026 was searched, with earlier foundational assessment research retained when conceptually necessary. Evidence was synthesised around assessment validity, engineering task vulnerability, authentic and process-based assessment, AI literacy and evaluative judgement, academic integrity and detection, feedback and grading, equity, and programme-level governance.  The literature indicates that neither blanket prohibition nor unrestricted adoption provides a defensible general solution. Authenticity alone is also insufficient, because realistic take-home tasks can remain highly susceptible to undisclosed AI influence. A more robust approach separates two complementary purposes: protected evidence of independent competence  in threshold and safety-relevant capabilities, and AI-integrated evidence of professional judgement in tasks where responsible tool use is itself an intended outcome. These forms of evidence should be triangulated through staged work, oral explanation, live demonstration, provenance, and programme-level assessment mapping. AI detectors are too unreliable and potentially inequitable to serve as stand-alone evidence of misconduct, while AI-assisted feedback and grading show promise but require human oversight, particularly for complex engineering work. The review proposes a Dual-Assurance Assessment Architecture as an evidence-derived organising model rather than a validated framework. Its central implication is that assessment reform should prioritise the validity of inferences about student capability, not merely the detectability of AI use.