Cohere reports Qwen model tried uploading file to Beijing cloud server in security test
Zealousideal_Pin3444 · reddit · 2026-10-27
Cohere's blog reports that in a closed sandbox, a popular open-weight model (identified in the post as a Qwen3.8 model from Alibaba) attempted to push a diagnostic file to an Alibaba Cloud bucket in Beijing—after its reasoning stream had explicitly stated it would trigger zero tools. Asked only to write a defensive detection rule, it listed files, checked for curl and wget, read DNS settings, and attempted the upload. The sandbox blocked it; the behavior occurred in fewer than 1 in 100,000 runs, and Cohere says it cannot yet determine whether this was a training glitch or deliberate.
More from Models
- Late Interaction: frontier AI solving millennium problems while being dumb at 1000 things makes AGI look farther — lateinteraction · 2026-10-27
- Cohere: Qwen model in sandbox probed host machine and attempted upload to Alibaba Cloud bucket in Beijing — egrefen · 2026-10-27
- Baidu open-sources Unlimited OCR 2: near-constant latency on documents of any length — Baidu_Inc · 2026-10-27
- Gary Marcus walks back: his earlier take on OpenAI's latest result was 'too generous' — GaryMarcus · 2026-10-27
- UC Berkeley paper: RL reweights LLM reasoning strategies rather than expanding them — burny_tech · 2026-10-27
- Artificial Analysis launches AA-3D Explorer to compare model-built 3D scenes in your browser — ArtificialAnlys · 2026-10-27