New Global Rules Ban AI Authors and Forged Citations in Medical Research

2026-05-25 global

Geneva, Monday, 25 May 2026.
Released today, new global medical guidelines strictly ban AI from being named as authors or generating citations, setting a crucial standard to protect data accuracy and human accountability.

Establishing Boundaries in Generative AI

On May 25, 2026, a new set of global consensus guidelines was published to govern the responsible use of generative artificial intelligence in medical research [1]. Prior to this unified framework, a review of AI policies across 15 major publishers, including the JAMA Network and Elsevier, highlighted significant inconsistencies [1]. The newly minted rules unequivocally prohibit listing AI as an author, generating fake references, or utilizing AI to create or manipulate primary research images [1]. This definitive stance aims to address persistent concerns regarding data leaks and AI masquerading as human researchers, ensuring that human authors remain fully accountable for scientific integrity [1].

The Push for Evidence-Based AI Tools

As medical publishing tightens its regulations against generative fabrications, the demand for evidence-based AI tools has surged [GPT]. One such platform is Consensus AI, an academic search engine that analyzes over 200 million peer-reviewed scientific papers sourced from databases like Semantic Scholar and PubMed [2][3]. Unlike standard predictive text models that require independent fact-checking to avoid hallucinations, Consensus AI grounds its responses directly in published evidence, covering publications spanning from the 1960s to May 2026 [2][3]. The platform continuously indexes new papers, typically incorporating peer-reviewed research within two to four weeks of publication [3].

Layered Risk Management in Clinical Settings

The push for AI regulation extends beyond academic publishing and directly into clinical workflows [4]. Artificial intelligence is already active in emergency departments through ambient documentation pilots, sepsis alerts, and staffing forecasts [4]. Following a consensus statement issued by the American College of Emergency Physicians on March 18, 2026, emergency physicians are now navigating how to safely integrate these tools [4]. The core challenge lies in differentiating low-risk administrative tools from high-stakes predictive models that actively influence patient disposition [4].

Global Disparities and Development Frameworks

Developing tailored digital health technologies (DHTs) for these clinical settings requires structured methodologies, a need recently addressed by the “Co-Develop-IT!” guideline published in the Journal of Medical Internet Research on May 22, 2026 [5]. Created by a multidisciplinary team of 10 researchers across four countries, this consensus-based guideline outlines an eight-phase iterative process for designing and evaluating DHT-enhanced training and rehabilitation concepts [5]. By introducing five preparatory contextual research phases before any generative co-development begins, the framework aims to bridge the evidence-to-practice gap and align the fast-paced expectations of industry partners with the rigorous demands of public research institutions [5].

Sources

Artificial Intelligence Medical Research