Biomedical Research Images Manipulated by Generative AI — Diagnostic Study of Detection Failures
OVERVIEWA diagnostic study of 104 western blot and subcellular xenograft tumor images reveals critical vulnerabilities in detecting generative AI-created forgeries. High-fidelity generative models produced forgeries that could materially alter study conclusions, yet 24 PhD-level expert reviewers could not reliably distinguish forgeries from authentic figures (mean accuracy 50.5%, SD 6.9%). The best-performing commercial AI detector achieved only moderate discrimination (AUC 0.790, 95% CI 0.695–0.885). This finding has profound implications for biomedical publishing and research integrity. The inability of expert human reviewers to detect sophisticated AI-generated images at better-than-chance rates suggests that current peer review processes are inadequate safeguards against AI-enabled scientific fraud. Publishers and journals must implement stronger forensic tools, institutional image provenance tracking, and revised disclosure standards for any computational image processing. The research community cannot rely on human visual inspection alone. ---
Subscribe to read full wiki analysis
Full wiki content — including source provenance, contradiction tracking, related pages, and analyst notes — is available to Subscriber members.
View Plans & SubscribeNew users get a 7-day trial — no credit card required.