The World Endoscopy Organization released a landmark international consensus statement in December 2025 that establishes a comprehensive legal and ethical framework for artificial intelligence applications in gastrointestinal endoscopy.
Published in Annals of Internal Medicine, the statement responds to rapidly accelerating AI adoption in endoscopic practice while simultaneously highlighting persistent gaps in how the technology should be governed, deployed, and monitored.
Developed through a two-round Delphi process involving 14 experts from 11 countries, the consensus emerged as part of the OperA project—Optimising Colorectal Cancer Prevention through Personalized Treatment with Artificial Intelligence—funded by the European Commission.
The experts agreed on 10 key statements spanning three interconnected domains: data governance, medicolegal implications, and equity and bias mitigation.
The Data Privacy and Governance Challenge
Data governance stands as the most fundamental concern addressed in the consensus statement. As AI algorithms for endoscopy generate and process vast amounts of patient information, questions about data ownership, privacy safeguards, and secondary use have become increasingly urgent.
The statement mandates that AI algorithms for gastrointestinal endoscopy must adhere to local information governance policies and robust data-protection regulations. All AI systems, whether commercial or developed for research purposes, should follow these protocols while establishing clear policies on data ownership and usage so patients understand how their information will be utilized.
Transparency regarding algorithmic modifications represents another critical dimension of data governance. Developers of AI tools must implement mechanisms to document and transparently report algorithm updates, modifications, and performance outcomes to both regulatory and clinical stakeholders.
These practices strengthen regulatory oversight, enable postmarket surveillance, and help clinicians understand how models change and perform in everyday clinical settings.
The challenge intensifies when considering endoscopy video data, which represents high-dimensional information at scale—recordings can reach 30 high-definition frames per second during procedures.
Traditional privacy protections like HIPAA deidentification have proven insufficient; multiple documented reidentification efforts using deidentified health information demonstrate the limitations of current frameworks. The consensus acknowledges that gastroenterologists, practices, and hospitals must increasingly navigate vendor data use contracts as the commercial value of endoscopy data rises.
Navigating Medicolegal Liability
The medicolegal landscape creates particular complexity for AI integration in endoscopy. When clinicians must rapidly balance AI-generated insights with clinical judgment in real-time settings, conventional boundaries of medical liability become blurred.
Critical questions remain unresolved: Where does liability rest if a clinician relies on an inaccurate computer-aided diagnosis (CADx) system? What risks emerge when an endoscopist dismisses an accurate AI report identifying inadequate mucosal inspection?
The consensus statement recommends that physicians and healthcare organizations ensure AI systems are used strictly according to manufacturer specifications. Medical societies and legal experts should provide clear guidance to mitigate liability concerns, particularly as semi-automated capsule endoscopy reading and more sophisticated CADx applications become widespread.
Before AI-driven automated report generation and AI-enabled quality metrics achieve widespread adoption, the statement emphasizes the necessity to evaluate their accuracy, understand clinical relevance, and clarify associated medicolegal implications.
The concern extends to automation bias—the tendency for clinicians to uncritically accept AI outputs. Research increasingly documents that exposure to AI-assisted detection can paradoxically reduce clinician performance.
Some studies show continuous AI exposure reduces adenoma detection rates when endoscopists subsequently perform standard colonoscopy without AI assistance, from baseline rates of 28.4 percent down to lower levels. This underscores that AI should function as a "second pair of eyes" providing reassurance or elevated confidence when aligned with clinical judgment, rather than serving as a substitute for physician expertise.
Confronting Equity and Algorithmic Bias
The consensus dedicates substantial attention to equity and bias, recognizing that AI systems can perpetuate or amplify healthcare disparities if training data lacks demographic representation.
Healthcare research has consistently demonstrated that AI tools yield biased outcomes when training datasets inadequately represent the populations they serve.
The statement recommends that AI algorithms for gastrointestinal endoscopy be trained and validated on datasets reflecting the race, ethnicity, and gender composition of populations they serve. Additionally, AI research and development initiatives should transparently report study population characteristics, enabling clinicians to assess the generalizability and equity of specific technologies.
Although individual AI models may not require stratification by race or ethnicity for every application, the consensus warns against defaulting to assumptions of irrelevance—such oversight risks perpetuating subtle but consequential inequities.
The practical concern manifests across geographic regions. An AI model trained predominantly on international populations may perform differently when applied to diverse local patient populations.
Research remains essential to determine whether AI adoption inadvertently exacerbates disparities through underrepresentation in training datasets or creates unequal access to AI technology across different healthcare settings.
Current AI Applications and Technical Landscape
The scope of AI implementation in endoscopy encompasses multiple functional categories. Computer-aided detection (CADe) systems identify and localize abnormalities during procedures, serving as an enhanced detection aid.
Computer-aided diagnosis (CADx) systems go further by characterizing lesions and assessing their clinical significance. Computer-aided quality assessment (CADq) systems monitor procedural quality, tracking withdrawal time adequacy, bowel preparation quality, and mucosal visualization completeness.
Current clinical evidence demonstrates measurable improvements in adenoma detection rates when AI assistance is employed. Meta-analyses examining 52 randomized controlled trials with over 50,000 patients show different AI-assisted interventions significantly improve detection compared with routine colonoscopy.
The ENDOANGEL model-assisted colonoscopy achieves the highest adenoma detection rate performance at 97.8 percent, while Endocuff-AI model-assisted colonoscopy demonstrates superior performance for sessile serrated lesion detection at 94.4 percent. However, for real-time polyp diagnosis using CADx systems, studies show minimal improvement in diagnostic sensitivity compared to clinician optical evaluation alone.
Implementation Challenges and Research Imperatives
The consensus statement identifies substantial research gaps requiring urgent attention. Most AI applications in endoscopy remain in preclinical or early clinical validation stages with important technical, regulatory, and ethical limitations unresolved.
Further research must evaluate real-world clinical impact beyond technical performance metrics, ideally linking AI adoption to meaningful patient outcome improvements rather than solely to technical accuracy measures.
A fundamental paradox complicates widespread adoption: while AI systems demonstrate superior performance in controlled settings, trust requires transparency, predictability, and understanding—characteristics often compromised by the "black box" nature of deep-learning algorithms.
The field lacks standardized approaches to interpret AI decision-making or adequately explain algorithmic recommendations to clinicians and patients.
Prospective implementation studies remain essential for understanding how AI adoption will transform gastrointestinal endoscopy practice, including workflow modifications, accountability structures, and evolving standards of care.
Research demonstrating AI performance across diverse populations and care settings will prove critical for avoiding potentially widened existing disparities.
Framework for Responsible Integration
Despite recognizing these challenges, the consensus statement provides an important starting framework for responsible AI governance in endoscopy. The recommendations represent neither prohibition nor uncritical endorsement, but rather structured guidance for addressing data governance, privacy, transparency, data ownership concerns, and risks of algorithmic bias.
Healthcare organizations implementing AI systems should develop clear internal policies, maintain transparent documentation of algorithm performance, ensure diverse dataset representation, and establish transparent informed consent processes explaining AI's role in patient care.
The statement acknowledges that different healthcare contexts—from resource-rich academic centers to resource-constrained settings—will require tailored approaches.
Regulatory bodies, professional societies, technology developers, and clinical practitioners must collaborate to develop implementation strategies reflecting local governance frameworks while maintaining core ethical principles.
The timing of this consensus proves significant given current AI implementation trajectories. Computer-aided detection for colonoscopy has been in clinical use in Europe since approximately 2019, in Asia since 2020, and in the United States since 2021. Ambient scribe technologies, automatic report generation, and numerous computer-aided diagnosis applications are advancing through development-to-deployment pipelines.
Establishing governance frameworks before widespread adoption becomes entrenched represents a critical opportunity for integrating privacy, transparency, and equity considerations into standard practice from inception rather than retrofitting these safeguards afterward.
The consensus statement ultimately reflects recognition that artificial intelligence's promise in gastrointestinal endoscopy—improved detection, reduced clinician workload, enhanced quality assurance—cannot materialize safely and equitably without robust governance structures.
Moving forward requires sustained collaboration among regulators, technology developers, medical societies, and clinical practitioners committed to ensuring AI integration benefits all patient populations while maintaining clinician expertise and accountability.
