IEEE Review Processes Detected All Skuse AI-Generated Papers Despite Zero Post-Generation Edits

Skuse ran an unmodified pipeline that executed experiments and produced full manuscripts without further intervention. Eliza Strickland confirmed the output passed initial plausibility filters yet failed downstream integrity review at IEEE. The experiment exposed that free-tier LLMs still insert detectable statistical artifacts that current research-integrity tools flag reliably.

Advanced models reduce those artifacts. Skuse’s trials showed free versions tripped basic detectors while paid frontier models produced fewer obvious markers. IEEE therefore faces an accelerating detection gap as commercial LLMs improve coherence and citation realism. Wikipedia’s WikiProject AI Cleanup reports parallel pressure with smaller volunteer capacity and no equivalent automated suite.

The operational result is sustained manual workload for gatekeepers. Publishers must now budget continuous retraining of classifiers and expanded human screening rather than static rules. Without those investments, the probability rises that AI-authored work enters the formal record and propagates through citation graphs.

Next threshold is 2026 model releases. If detection false-negative rates exceed 15 percent on blind tests, IEEE and similar bodies will require mandatory author tooling disclosure and compute provenance logs.