Alibaba—Investigation found Alibaba's Qwen AI model embeds CCP-aligned censorship and positive-framing bias on sensitive political topics
A CBS News investigation and a China Media Project analysis found Alibaba's open-weight Qwen models return canned CCP-aligned responses or refusals on topics like Taiwan, Tiananmen Square, and Xinjiang, and were found to follow instructions to describe China's human-rights record using only positive framing while listing both positives and negatives for other countries. Independent benchmarking cited in the reporting found roughly 90% of politically sensitive prompts triggered censored or biased responses in base Qwen models. Qwen has been downloaded over 3 billion times and is used inside products at companies including Airbnb and Uber, spreading the bias into international business tools. As of the reporting, Alibaba had not publicly addressed the findings.
Scoring Impact
| Topic | Direction | Relevance | Contribution |
|---|---|---|---|
| Authoritarian Compliance | +toward | primary | -1.00 |
| Misinformation | +toward | secondary | -0.50 |
| Overall incident score = | -0.643 | ||
Score = avg(topic contributions) × significance (high ×1.5) × confidence (0.57)
Evidence (1 signal)
CBS News investigation finds Qwen AI censors and spins politically sensitive topics for China
CBS News tested Alibaba's free Qwen AI model and found it refuses or gives canned CCP-aligned answers on topics including Taiwan, Tiananmen Square, and Xinjiang, while applying positive-only framing to China's human rights record.