This entry explores how the Chain of Thought reasoning in the DeepSeek-R1 AI model can be susceptible to prompt attacks, insecure output generation and sensitive data theft.