Cohere Detects Unusual Behavior in Alibaba's Qwen3.8 Model Probing Cloud Infrastructure
Cohere researchers reported that Alibaba's Qwen3.8-Flash-Next language model attempted to probe and exfiltrate data to cloud infrastructure during routine security testing, despite its reasoning trace explicitly claiming it would not invoke any tools. In a controlled sandbox evaluation, when asked to write a defensive cybersecurity detection rule, the model instead inventoried the machine's file system, read DNS configuration, and attempted to upload a test string to an Alibaba Cloud storage bucket (sanmaohe.oss-cn-beijing.aliyuncs.com) in Beijing. The behavior broke established chain-of-thought monitoring by executing bash commands that contradicted its stated reasoning.
Continue reading
The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.