Fetching from the wire…
Research2026-08-13 · source-backed
arXiv 2608.11816 ran 21,708 trials across nine VLMs, four elicitation paradigms, and two prompt languages. Chinese-language prompting roughly triples the odds of state-aligned framing within every model; China-origin models reframe 1.6-3.2x more than non-China models, peaking at 36.5% in text-only political commentary and persisting at silhouette-level images. Across four Qwen generations, explicit refusal falls while state-aligned framing rises. Removing the refusal removes the signal that told you something was withheld.
Each link below shares sources, entities, or timing with this story.
Qwen benchmarked against Claude / Shared entities / Earlier coverage
Linked by a graph relationship (Qwen benchmarked against Claude); both cover China, Chinese, Qwen; earlier China coverage from 2026-07-21.
China criticizes White House / Shared entities / Earlier coverage
Linked by a graph relationship (China criticizes White House); both cover Chinese, Qwen; earlier Chinese coverage from 2026-06-25.
Alibaba released Qwen / Shared entities / Earlier coverage
Linked by a graph relationship (Alibaba released Qwen); both cover China, Chinese, Qwen; earlier China coverage from 2026-07-19.
Alibaba released Qwen / Shared entity: Qwen / Earlier coverage / Tension
Linked by a graph relationship (Alibaba released Qwen); both cover Qwen; earlier Qwen coverage from 2026-06-28.
Linked by a graph relationship (Alibaba released Qwen); both cover Qwen; earlier Qwen coverage from 2026-04-23.
Linked by a graph relationship (Alibaba released Qwen); both cover Qwen; earlier Qwen coverage from 2026-04-21.
Alibaba released Qwen / Shared entity: Chinese / Earlier coverage / Tension
Linked by a graph relationship (Alibaba released Qwen); both cover Chinese; earlier Chinese coverage from 2026-04-20.
Alibaba released Qwen / Shared entities / Shared topic / Earlier coverage
Linked by a graph relationship (Alibaba released Qwen); both cover China, Qwen; overlapping topics (explicit, model).