Can Research Integrity Be Automated? The Paradox of AI Governing AI in Science

Abstract

Artificial intelligence is increasingly embedded in scientific work, supporting writing, analysis, interpretation, and decision-making. At the same time, AI is also being deployed as a governance tool to monitor misconduct, check originality, and enforce procedural compliance. This dual expansion creates a fundamental paradox: the same technologies that may weaken research integrity are also expected to protect it. The study addresses this paradox by asking whether research integrity can be meaningfully automated when AI increasingly mediates both scientific production and its oversight. To examine this problem, the article develops a dual-automation integrity model that formalizes the relationships between AI automation in scientific work, AI automation in integrity oversight, human epistemic accountability, procedural compliance, and substantive research integrity. The model introduces two synthetic indicators (the compliance-integrity gap and the recursive governance paradox index) and illustrates their behavior through two contrasting numerical scenarios. The findings show that automation does not have a uniform effect on research integrity. Under conditions of balanced augmentation, AI can support scientific work while preserving strong human accountability and substantive integrity. The model contributes a conceptual and analytical framework for understanding the limits of automated integrity governance and for guiding future debate on responsible AI in science.