A critical-severity vulnerability in the open source AI gateway LiteLLM, identified as CVE-2026-42208, was exploited just days after it was publicly disclosed, allowing attackers to access sensitive database tables. This vulnerability, which has a CVSS score of 9.3, stems from an SQL injection error in the proxy API key verification process, where caller-supplied values were not properly separated from the database query. After the advisory was published on April 20, attacks exploiting the flaw began occurring 36 hours later, targeting database tables that contained API keys and credentials. LiteLLM’s maintainers have since released version 1.83.7 to patch the vulnerability by ensuring proper parameterization of database queries and have advised users to either update or disable error logs to mitigate further exploitation.

Sysdig: Sysdig is a cloud-native security platform providing runtime visibility, threat detection, and response for cloud environments. Founded by creators of open-source tools like Falco, it emphasizes real-time monitoring of containers and cloud workloads. Sysdig researchers detected and analyzed the targeted exploitation of the LiteLLM vulnerability, highlighting the attacker’s precise schema enumeration techniques.
LiteLLM: LiteLLM is an open-source proxy server and Python library that serves as a unified gateway for accessing numerous large language model APIs through an OpenAI-compatible interface. It handles features like authentication, load balancing, and fallback routing across various LLM providers. In this incident, a pre-authentication SQL injection vulnerability in the proxy’s API key verification allowed attackers to access sensitive database tables shortly after disclosure.

Resolution: LiteLLM maintainers patched the flaw by passing values as separate parameters in queries, advising users to update or disable error logs.
Attack Pattern: Exploits used crafted Authorization headers accessible via the proxy’s pre-auth error-handling path to read database tables with API keys and credentials.
Vulnerability Mechanism: The SQL injection occurs in a database query during proxy API key verification where caller-supplied values were directly included without parameterization.