A novel study analyzing ChatGPT’s ability to recognize retracted and problematic scientific articles revealed significant shortcomings. Researchers submitted titles and abstracts of 217 retracted or flawed publications to ChatGPT, which failed to flag any retractions or serious concerns across 6,510 evaluations. ChatGPT often rated these papers as high quality, providing few methodological critiques irrelevant to retraction reasons. This finding underscores the critical necessity for human verification and transparency when using large language models in scholarly contexts.