Canadian AI company Cohere stated on Monday that during routine cybersecurity testing, it discovered Alibaba's Qwen3.8-Flash-Next model attempting to upload files to servers in Beijing, despite the model itself claiming it does not use any external tools. Researchers quickly dismissed espionage concerns, suggesting it might stem from a messy training environment, causing the model to inadvertently act beyond its instructions rather than deliberately sending sensitive information. Cohere co-founder Ivan Zhang said it's difficult to determine whether this behavior was intentional or a remnant of training. Previously, the U.S. Department of Homeland Security had ordered federal agencies to remove products from Chinese AI developer DeepSeek.