Generative artificial intelligence systems now draft essays, analyze data, and produce multimodal artefacts with near-human proficiency, placing coursework validity at risk across disciplines. Existing safeguards either detect misconduct after the fact, rely on secure-exam settings that constrain authentic tasks, or ask lecturers to perform manual redesigns that are difficult to scale. This paper presents AIMA (AI-based Assessment Integrity Management Assistant), an automated platform that integrates two complementary tracks: (1) a static analysis that scores assignment briefs on eight linguistic and structural factors linked to generative-AI susceptibility, and (2) a dynamic, time-boxed “robot student” powered by agentic AI that attempts the task and is auto-graded against the rubric. The fused evidence produces a 0–100 vulnerability index (AIMA Score) plus sentence-level revision suggestions. A design-science research methodology guided development, and a formative pilot evaluated four authentic briefs (two in Marketing, two in Data Science/AI) authored by the researchers. AIMA processed each brief in under two minutes; lecturers adopted between one and three suggested edits, reducing vulnerability scores from Amber or Red to Green within fifteen minutes. System Usability Scale ratings averaged 83.5, indicating excellent acceptance. Although the sample is small, findings demonstrate that agentic AI can provide rapid, actionable support for assessment design without disclosing sensitive implementation details. Future work will expand empirical testing, add explainability layers, and track student outcomes over time.