Routine use of artificial-intelligence software during colonoscopy was linked to a measurable loss of diagnostic skill once the technology was withdrawn, according to a study published in Lancet Gastroenterology & Hepatology. Researchers reviewed 1,443 standard, non-AI-assisted colonoscopies performed at four Polish centres participating in the ACCEPT trial. The adenoma detection rate fell to 22.4% in the three months after clinicians had routinely worked with AI support, down from 28.4% in the three months before the software was introduced—an absolute decline of six percentage points, or roughly 20%. Exposure to AI remained a significant factor in a multivariable analysis that adjusted for patient age and sex. The authors say the findings are the first real-world evidence that clinical AI can ‘deskill’ experienced practitioners, raising concerns about over-reliance on automation as hospitals accelerate deployment of diagnostic algorithms. While earlier trials showed AI boosts detection when active, the new data suggest outcomes could worsen if systems fail or are unavailable. The team, whose work was funded by the European Commission and Japan’s science-promotion agency, called for further research and strategies to ensure clinicians maintain core competencies as the technology spreads across medicine.