A critical Cypher injection vulnerability in the Graphiti framework allows authenticated attackers to execute arbitrary database queries, potentially leading to unauthorized data access or modification.

This vulnerability is a Cypher injection flaw (CWE-943) occurring within shared search-filter construction. It allows an authenticated user, or an attacker leveraging prompt injection against an LLM client, to inject malicious input into Cypher label expressions.

The ability to perform Cypher injection allows attackers to bypass intended query logic, leading to the potential exposure or modification of sensitive data stored within the graph database. Given the CVSS score of 8.1, this vulnerability poses a significant risk to the integrity and confidentiality of AI-driven temporal context graphs. Organizations relying on Graphiti for LLM orchestration are at particular risk, as the flaw can be triggered indirectly through prompt injection.

Immediate Action: Update the Graphiti framework to version 0.28.2 or later to apply the necessary fix for Cypher query construction.

Proactive Monitoring: Review application logs for unusual query patterns or unexpected input values within the SearchFilters.node_labels parameter.

Compensating Controls: Implement strict input validation on all data passed to LLM-orchestrated functions and ensure that backend database permissions follow the principle of least privilege.

Public Exploit Available: Yes, a public proof-of-concept repository exists on GitHub.

Due to the severity of potential data compromise and the ease with which this injection can be triggered, organizations must prioritize upgrading to version 0.28.2 immediately. Security teams should verify that their deployments are not using vulnerable backend configurations and should audit any existing integrations between LLM clients and the Graphiti MCP server to ensure no malicious vectors remain accessible.