This critical SQL injection vulnerability in BerriAI LiteLLM is confirmed to be actively exploited in the wild and poses a severe risk of unauthorized access to proxy credentials.

The vulnerability exists because the proxy improperly neutralizes special elements in database queries during API key verification. An unauthenticated attacker can exploit this by sending a malicious Authorization header to any LLM API route, which reaches the vulnerable query through error-handling paths.

Successful exploitation allows an attacker to read and potentially modify sensitive data within the proxy database, including managed API credentials. Given the 9.5 CVSS score, this represents a critical threat to the confidentiality and integrity of AI infrastructure. The inclusion of this CVE in the CISA Known Exploited Vulnerabilities catalog underscores the high urgency for immediate remediation.

Immediate Action: Upgrade all instances of LiteLLM to version 1.83.7 or later immediately to resolve the injection flaw.

Proactive Monitoring: Review proxy access logs for anomalous Authorization headers or unexpected database query patterns that deviate from standard API interaction.

Compensating Controls: Deploy a Web Application Firewall (WAF) with rules configured to detect and block SQL injection patterns within HTTP headers.

Public Exploit Available: Yes, a Metasploit module and Nuclei templates exist.

The critical nature of this vulnerability, combined with confirmed active exploitation, necessitates immediate action. Organizations must prioritize patching to version 1.83.7 to prevent potential compromise of sensitive AI credentials and proxy data.