An IEEE Spectrum writer used free, off-the-shelf AI tools to generate a convincing research paper about work that was never performed. Then IEEE Access's managing editor agreed to war-game whether the fake could slip past a real journal's defenses, testing how easily AI slop enters science.
By 2025, an estimated 57 percent of published academic articles showed detectable evidence of AI influence, up from 12 percent in 2023. Yet only one in 40 papers with statistical signs of AI use formally disclosed it. This experiment shows how little expertise it takes to pollute the scientific record — the record doctors and policymakers rely on.
✦ If faking science is this easy, verifying it becomes everyone's job too.