• [R] Illusory Safety: Redteaming DeepSeek R1 and the Strongest Fine-Tunable Models of OpenAI, Anthropic, and Google http://far.ai/post/2025-02-r1-redteaming/ 9 comments 5/2/2025 machinelearning