Researchers Propose Formula to Predict When AI Chatbots May Turn Dangerous — 09.10.2026

Researchers from George Washington University published a paper examining whether the time and cause of an AI chatbot going rogue can be predicted and, if predicted, prevented. They developed a mathematical formula to estimate the tipping point, represented by the number of good outputs before the first undesirable output appears.