This study evaluates whether Large Language Models (LLMs) can accurately interpret clinical trial results when presented with biased or spun medical literature.

• LLMs are susceptible to being misled by spin in medical research abstracts
• Models often fail to identify discrepancies between actual data and authors' interpretations
• This vulnerability raises concerns about using LLMs for medical information processing
• The findings highlight the need for advanced training of AI models to detect and resist misleading scientific reporting

Why it matters: As LLMs increasingly influence healthcare decisions, their inability to detect spin in medical literature could propagate misinformation to clinicians and patients, potentially affecting treatment choices and outcomes.

Caught in the Web of Words: Do LLMs Fall for Spin in Medical Literature?