R2R version 3.6.6 contains a SQL injection flaw that allows an attacker to embed SQL predicates into the retrieval search query by manipulating the filter key parameter. The application directly interpolates these keys into the SQL WHERE clause without using parameterization or escaping, enabling both boolean‑based and time‑based attacks to exfiltrate data. This flaw can be triggered by unauthenticated requests, thereby exposing sensitive database contents to any external actor. The impact is primarily the compromise of data confidentiality rather than denial of service or code execution.

The affected product is SciPhi‑AI R2R running the 3.6.6 release. Any deployment of this version without additional safeguards is susceptible to the described injection vector via the retrieval search endpoint.

The CVSS score of 8.7 classifies this as a high severity vulnerability. No EPSS score is available, and the issue is not listed in the CISA KEV catalog. Attackers can exploit the flaw without authentication by sending a crafted filter key to the search API; the lack of input sanitization or parameterization makes the attack straightforward, while the absence of a known exploit in the wild is currently unknown. Given the high severity and potential for sensitive data loss, rapid remediation is strongly advised.