Questionable research practices in the use of AI

Artificial Intelligence is rapidly being integrated into research workflows, but its use is not free from risks and questionable practices. This presentation explores the emerging spectrum of questionable research practices (QRPs) in AI, ranging from fundamental technical tricks in machine learning—such as data contamination, cherrypicking, and misreporting—to the misuse of AI in scientific writing, peer review, and grant applications. Specific cases include prompt injections in peer review, breaches of confidentiality through AI-assisted reviewing, “AI-giarism” (plagiarism with AI-generated text), fabricated citations, and the uncritical acceptance of AI outputs. These practices pose significant risks to research integrity, data protection, confidentiality, and trust in science. The talk highlights concrete examples, recent studies, and guidelines from organizations such as UKRIO and the DFG, and concludes with reflections on how researchers and institutions can safeguard integrity while responsibly integrating AI into their work.