Science Behind a Paywall: Restricted Access Limits the Promise of Artificial Intelligence

Key Points Paywalls restrict AI access to approximately 50% of full‐text scholarly articles, limiting training on high‐quality data essential for scientific accuracy. Lack of full‐text access for AI risks a self‐reinforcing cycle of low‐quality outputs that potentially erode the integrity of scholarly publishing. Paywalls fund vital publishing infrastructure and are legally protected, making paywall removal unsustainable. Emerging licencing partnerships together with RAG offer practical ways to grant AI secure access while sustaining publishers. Urgent action to bridge this gap is needed before AI‐generated synthetic data dominates.