Attackers Inject Ransom Demands Into AI Chatbot Knowledge Bases Through Unprotected Vector Databases

Researchers have documented the first attack in which malicious actors injected ransom demands directly into Chroma vector databases, causing AI chatbots to return extortion messages to users querying the compromised knowledge base. The incident exploits the widespread deployment of Chroma—an open-source vector database downloaded approximately 14 million times monthly and used as the default store in LangChain and LlamaIndex—often without authentication or access controls.