The advent of Generative Artificial Intelligence (GenAI) poses a significant threat to the integrity of scientific literature. While Open Science is a crucial safeguard against fraudulent or unethical activities, it inadvertently fuels AI-generated fraud. Open access datasets, especially in health and medicine, are being exploited by paper mills and other bad actors, leading to a surge in low-value or misleading research. This has detrimental consequences, including misallocation of funding, distorted assessment metrics, and reduced trust in the scientific literature. The scientific community and publishers are actively addressing these issues, but the challenge remains. The problem lies in the open access policy, which, while promoting equitable data access and improved reproducibility, also enables AI-assisted fast churn science. The UK Biobank, for instance, provides open access to de-identified data, but a pilot audit revealed a concerning trend: 7% of research papers did not report or appear to have a valid UK Biobank application number, and 25% exhibited substantial hypothesis drift. This highlights the vulnerability of open-access data sources. The NHANES dataset is another example, where some publishers no longer accept submissions based on open-access public health datasets. This demonstrates the real cost of not exerting preventive control over data usage. The solution lies in a balanced approach. While unrestricted open access promotes ethical and scientific benefits, it needs to be complemented with better disclosure processes and publication checks. These measures would include de-registering applications with significant hypothesis drift and focusing on legitimate data acquisition. By reasserting the balance between Open Science and research integrity, we can safeguard the credibility of research based on open access data sources.
