Introducing NOVA

NOVA is an open-source prompt pattern matching tool for hunting malicious or policy-violating prompts in generative AI systems, combining keyword/regex detection, semantic similarity via embeddings, and LLM-based evaluation within YARA-like rules and conditions. The article details rule structure (meta, keywords, semantics, LLM, condition), configuration and CLI usage, and outlines detectable behaviors such as prompt injection, jailbreaking, malware/scam generation, reconnaissance, and harmful content—aimed at teams operating or monitoring AI systems and analyzing prompt logs.