DeepSeek can be gently persuaded to spit out malware code

Tenable researchers demonstrated that the DeepSeek R1 generative AI's guardrails can be bypassed via careful prompting to generate working C++ keylogger and rudimentary ransomware code that, with manual fixes, can run and hide artifacts; the article places this finding alongside examples of attacker-built LLMs (WormGPT, FraudGPT, etc.) and national warnings about AI's potential to aid offensive cyber operations.