CVE-2024-8309

An attacker can inject malicious instructions through the prompt injection mechanism, which are then passed without proper sanitization to queries executed by the GraphCypherQAChain class. Lack of proper input validation (CWE-89, CWE-74) allows embedding arbitrary commands in the context of graph database queries. As a result, an attacker can perform operations to create, modify, and delete nodes and relationships in the database without required authorization.

An attacker can conduct unauthorized exfiltration of sensitive data, delete all data causing a DoS attack, breach data isolation in multi-tenant environments, and compromise database integrity by creating, updating, or deleting nodes and relationships.

Apply patches available from the vendor according to references — fix available in the GitHub repository under commit c2a3021bb0c5f54649d380b42a0684ca5778c255

langchain-ai/langchain version 0.2.5, GraphCypherQAChain component