<p>Academic institutions and governance bodies are implementing rules that restrict the use of AI tools in research. This paper develops a formal interest-group theory of AI-tool governance in science, modeling such rules as the endogenous outcome of organized conflict within the scientific community over policies adopted by journals, funders, departments, and professional associations. Scientists differ in how much they gain from AI tools, while rank-relevant rewards, such as publication opportunities, grants, promotions, prizes, and professional recognition, depend partly on relative standing. Even when AI tools raise aggregate knowledge production, scientists with smaller AI-related productivity gains can invest resources to secure restrictions that compress rivals’ rank advantages. The equilibrium likelihood of restriction, lobbying expenditures, and rent dissipation are characterized in closed form in a Tullock contest framework. The model yields a welfare wedge: because rank-based rewards are privately valuable and partly positional, the chosen rules can be excessively restrictive relative to the social optimum. The mechanism is most relevant where governance bodies retain substantial discretion over rule-setting and where scientists cannot easily avoid restrictive policies by shifting to other venues. Restrictions are more likely when disclosure, sanction, or reputational concerns about AI use are more prominent, and when scientists favoring stricter rules are better represented in rule-setting bodies or better able to coordinate pressure for restriction. When scientists’ average net private gain from permissive AI use is positive, restrictions are also more likely when rank incentives are steeper and when AI increases the relative-performance advantage of high-complementarity scientists. The analysis yields testable predictions that distinguish this interest-group account from a public-interest account of AI-tool governance.</p>