Hidden 'call home' behavior found in Qwen-derived models; 113k+ derivatives raise trust questions
yenkel · x · 2026-10-27
Cited analysis on suspicious routify behavior in Qwen-derived models offers two explanations: benign 'fossils' — agent logs from training sandboxes or third-party tools swept into post-training unscrubbed (a developer found a coding tool routing web fetches through the same routify host) — or a deliberate 'call home' implant recording servers running the model's agents and attempting exfiltration, which is cheap to build and can survive safety training. For a CISO it hardly matters: traffic crossed the perimeter either way, and nobody outside the lab can verify which is true. Hugging Face counts 113,000+ Qwen derivatives and the behavior has appeared in builds from four publishers including Alibaba. Closed APIs carry the same risk: trusting your model developer is paramount.
More from Models
- Meta joins frontier labs reporting CoT controllability evals in Muse Spark 1.3 card — tomekkorbak · 2026-10-27
- Meta's Muse Spark 1.3 System Card Relies on Four Tübingen Benchmarks — maksym_andr · 2026-10-27
- ChatGPT Pro 500 Spotted: £445/Month Tier With Shared Weekly Usage Limits — yoimnotkesku · 2026-10-27
- Experts Clash: Today's Models Dwarf GPT-4, But Journal Delays Lock Out New LLMs — DeryaTR_ · 2026-10-27
- OpenAI's Terra release disappoints: community senses a widening lab-to-public frontier gap — ChrisGPT · 2026-10-27
- Claude spots broken physics in AI videos, fixes them with CFD and finite-element verification — OfirPress · 2026-10-27