• LLMs Corrupt Your Documents When You Delegate, by Philippe Laban, Tobias Schnabel and Jennifer Neville (Arxiv, 17/04/2026) https://arxiv.org/html/2604.15597v1 “In a Microsoft Research study, researchers simulated long delegated workflows requiring extensive document creation and editing across 52 professional domains, including coding, crystallography, and music notation. After around twenty interactions, AI models significantly degrade the documents. Even the best-performing models corrupt an average of ~25% of the content by the end of long sessions. They alter figures in financial statements, make words disappear…”

 

• Certifying humanity: Labeling content amid AI flood, by Slobodan Kovrlija, an article published on 27 January 2026 on the DiploFoundation website: https://www.diplomacy.edu/blog/certifying-humanity-labeling-content-amid-ai-flood/ It explores the issues and initiatives surrounding the labeling/certification of human-produced content (guaranteed to be AI-free). It notably discusses the arguments for and against such approaches, including concerns in Global South countries, where locally produced content may be subject to standards defined elsewhere, or where verification infrastructures are unevenly available.