arXiv:2307.06865

System prompts in LLMs are not secure, researchers demonstrate

System prompts in large language models are not a secure boundary and should not be relied upon for security, according to a dev.to article. Researchers have demonstrated various methods, including direct questioning an…