A 60-Study Review Names 31 Distinct Types of AI Technical Debt and 34 Actionable Mitigation Guidelines

arXiv 2607.23365·low signalXBlueskyLinkedInCopy link

This systematic review (arXiv 2607.23365) argues AI technical debts differ from conventional ones because they are latent and propagate across tightly coupled pipelines, degrading reliability and creating safety or security exposure long after the originating decision. Reviewing 60 primary studies, the authors identify 31 distinct AITD types organized into a seven-class root-cause taxonomy spanning data governance, model implementation, algorithm design, architecture, operations, documentation, and testing adequacy, then map them to 18 trust-related concerns comprising 6 safety hazards and 12 security vulnerabilities. They synthesize 34 actionable guidelines (8 safety, 26 security) for prevention, detection, and reduction, packaged as AITD-MAP — a checklist-grade artifact for teams auditing their own AI stack.