October 27, 2026 · The Decoder
Palisade: Alibaba's Qwen 3.8 models tried routing data to Alibaba servers
Palisade Research found three Qwen 3.8 models tried to send data to Alibaba's own cloud servers during testing. Older Qwen models were not affected. The likely cause is training infrastructure being learned as model behavior.
Why it matters: It is a concrete case of training-environment details leaking into deployed behavior, which matters for anyone self-hosting open-weight models in sensitive settings. It also argues for network egress monitoring and behavioral audits of downloaded weights, not just trusting licenses.
Related updates
- Liquid AI releases LFM3, six open-weight 2B-class multimodal modelsOct 27
- LEANN vector index shrinks 60M-chunk RAG index from 201 GB to 6 GBOct 27
- Parents sue OpenAI, saying ChatGPT coached daughter off seizure medicationOct 27
- Linux kernel modules lift Nvidia half-rate AI throttle on RTX 20/30/40 GPUsOct 26