The WeKnora framework contains a SQL injection vulnerability triggered by the Agent service when database queries are executed after processing user prompts. The flaw stems from insufficient backend validation of queries, allowing an attacker to craft prompt-based bypasses that remove or alter mandatory restrictions. Successful exploitation lets the attacker read sensitive data stored on the target server and database, compromising confidentiality. This weakness is a classic SQL Injection flaw (CWE‑89).

The vulnerability is present in Tencent's WeKnora software prior to version 0.2.5. Any deployment that has the Agent service enabled and is running a pre‑0.2.5 build is at risk. The affected product is the WeKnora LLM‑powered document understanding platform; even a single node can leak data if the Agent component is enabled.

The CVSS base score is 5.6, indicating moderate severity. The EPSS score is less than 1 %, suggesting a low likelihood of widespread exploitation at this time. The vulnerability is not listed in CISA’s KEV catalog. Attackers would need remote access to the Agent service API and the ability to supply crafted prompts; thus the probable attack vector is remote, via user prompt submission, and is inferred from the description that the flaw is triggered by prompt‑based bypass techniques.