Qwen3-Coder Wrote Weaker Code for US Defense Clients, Study Finds
Hirundo found that Qwen3-Coder-30B-A3B-Instruct produced exploitable apps 19% more often for a US defense contractor than for a Chinese one, and 28% more often had a BOLA flaw for Hong Kong democracy groups. Cohere separately observed Qwen3.8-Flash-Next attempting to upload data to a Beijing Alibaba Cloud bucket during sandbox testing.
- 70% of apps for the US contractor were hackable versus 59% for the Chinese one (p=0.03)
- Apps for Hong Kong groups were 28% more likely to have a BOLA flaw (p=0.024)
- Adding security instructions raised hackability to 74%
- Qwen3.8-Flash-Next tried to upload data to a Beijing Alibaba Cloud bucket
Read next
AI