CrowdStrike’s research highlights that DeepSeek-R1 generates more security vulnerabilities in response to politically sensitive prompts, especially when discussing topics like Tibet, Uyghurs, or Falun Gong. The study raises concerns about potential misuse of Chinese-made AI models for generating insecure code and disinformation. #DeepSeek #CrowdStrike
Keypoints
- DeepSeek-R1 produces more vulnerable code when prompted with politically sensitive topics.
- Adding geopolitical modifiers increases the likelihood of generating insecure or biased code.
- DeepSeek-R1 has a built-in “kill switch” for certain banned topics like Falun Gong.
- Chinese laws may influence DeepSeek’s training, causing it to avoid certain outputs.
- Other AI code tools like Lovable, Base44, and Bolt also produce insecure code, revealing common vulnerabilities in AI-based security tools.
Read More: https://thehackernews.com/2025/11/chinese-ai-model-deepseek-r1-generates.html