Australian firms warned on AI risks after DeepSeek generated more flaws on sensitive topics.

Australian companies are being warned to proceed cautiously with foreign AI tools after cybersecurity firm CrowdStrike found China’s DeepSeek AI model generated significantly more security flaws in code when prompted with politically sensitive terms like Tibet, Taiwan, Falun Gong, and Uyghurs. The model was 50% more likely to produce vulnerable code—such as missing session management—on these topics, while delivering secure code for neutral requests. In some cases, it refused to respond, suggesting a possible built-in "kill switch." These findings, the first of their kind, raise concerns about ideological influences compromising AI safety and reliability. The research comes as Australia prepares to launch its AI Safety Institute and global scrutiny of AI governance intensifies.