One-Way Moves? Affiliation Trajectories of Iranian Artificial Intelligence Researchers Relative to Regional Peers, 2010–2025

Purpose. Concern about the emigration of Iranian scientists is widespread, but evidence on artificial intelligence (AI) researchers rests on small samples of faculty members and does not benchmark Iran against comparable systems. We ask whether Iranian AI researchers show a distinctive pattern of foreign-only affiliations after 2018, when they move, and where they go. Design and methods. We built cohorts of authors with at least two AI articles carrying a home-country affiliation in 2010–2017 (Iran n = 1,826; Türkiye 784; Egypt 436; Pakistan 303; Malaysia 844) from OpenAlex and followed their recorded institutional affiliations to 2025. Each author was classified, in 2019–2025, as home-only, dual (home and foreign) or foreign-only. We estimate adjusted logit models, author-year difference-in-differences and event-study models with country and year fixed effects and author-clustered errors, and report destination concentration, productivity selectivity and sensitivity to comparator choice and profile filters. Findings. Among authors still publishing in 2019–2025, 7.5% of Iranian researchers had foreign-only affiliations (95% CI 6.3–8.9) versus 7.1% for Türkiye, 4.9% for Egypt and 4.3% for Pakistan. Adjusted for baseline productivity, cohort entry and visibility, the odds were 1.42 times those of the Türkiye–Egypt–Pakistan pool (95% CI 1.05–1.93), but not different from Türkiye alone (1.20; 0.84–1.71), and lower than a pool that includes Malaysia (0.83; 0.65–1.07), whose home cohort contains many foreign students. Iranian researchers were far less likely to hold dual affiliations (odds ratio 0.53; 0.45–0.62), which indicates one-way rather than circular moves. Relative to Türkiye and Egypt the divergence was absent until 2022 and appeared in 2023–2025 (+3.0, +4.9 and +3.4 percentage points; difference-in-differences after 2021: +3.1 pp, 95% CI 1.2–5.1; pre-trend slope borderline, p = 0.043). With author fixed effects the post-2021 estimate fell to +1.6 pp (95% CI −0.3 to 3.5), was +2.0 pp (0.1–3.9) among authors visible in both follow-up periods, and vanished (+0.3 pp; −2.3 to 2.8) among authors visible in every year 2019–2025, so the late divergence is fragile and may partly reflect a 2022 coverage discontinuity in OpenAlex. Relocators concentrated in North America (United States 27.0%, Canada 21.7%; 48.7% versus 25.3% for peers). More productive researchers were not more likely to relocate (interaction p = 0.71). Originality. The study replaces anecdote and small faculty samples with a comparative, author-level design, shows that the signal is modest, late and comparator-dependent, and reports where affiliation-based proxies cannot support claims about emigration.