DeepSeek-R1 Exposes a New AI Weakness: Security Degrades With Ideological Triggers
CrowdStrike’s analysis shows DeepSeek-R1’s code security collapses when prompts include politically sensitive terms like Tibet or Falun Gong. The failure isn’t a bug; it’s alignment leaking into engineering. And it signals a broader shift: AI is already fragm…
Key Takeaways <ul><li>CrowdStrike found DeepSeek-R1’s code security collapses when politically sensitive keywords are present, even when those words have nothing to do with the task. Vulnerability r… [+9117 chars]