Detecting AI-Completable Assignments: An Explainable Generative AI Framework for Assessment Governance

As generative AI becomes embedded in student academic workflows, many traditional assignments can now be completed by general-purpose language models with minimal student-specific reasoning, process evidence, or contextual engagement. This paper introduces the concept of AI-completable tasks as a diagnostic construct for assessment governance in higher education. Rather than treating AI use primarily as a post-submission misconduct problem, the paper reframes the challenge as one of assessment design alignment. A faculty-side, generative AI evaluator is presented to analyze assignment prompts, simulate AI substitution risk, assign an AI-resistance score, identify structural vulnerabilities, and recommend targeted redesign strategies. Using an artifact-based pilot that reviewed approximately 100 assignments, the study reports representative computing-focused pre- and post-intervention cases to examine how design changes alter AI-completability. Results suggest that modest structural revisions, including contextual grounding, process evidence, student-specific artifacts, and trade-off justification, can reduce evaluator-assessed AI substitution risk while preserving original learning objectives. The paper contributes an explainable AI-supported framework for faculty-centered assessment governance and positions generative AI as a proactive design aid rather than merely a detection or enforcement challenge.