AI Debt: Verification Burdens, Rework Externalities, and the Hidden Costs of Asymmetric Generative-AI Adoption

Generative AI is usually evaluated through the productivity of the immediate user: faster drafting, more output, lower search cost, and apparent acceleration of professional workflows. This article argues that such evaluation is incomplete. When AI-generated artifacts are under-verified, the costs are often displaced onto downstream recipients: judges, teachers, reviewers, managers, clients, colleagues, and administrative staff who must check, correct, contextualize, or reject low-quality AI output. Building on the software-engineering metaphor of technical debt, the article introduces the concept of AI debt: the accumulated obligation to remediate AI-generated artifacts that were produced faster than they were validated. AI debt has a principal, an interest rate, a debtor, and frequently a different burden bearer. The article develops a conceptual framework, reviews evidence from legal practice, workplace knowledge work, education, academic publishing, and professional services, and proposes governance mechanisms inspired by technical-debt management: debt registers, definition-of-done rules, review gates, automated citation tests, amortization sprints, receiver-side rights, and accountability allocation. The central claim is that responsible AI adoption must be measured at the workflow and system level, not only by individual user time savings.