Skip to content
Threat Feed
critical threat exploited PoC updated

LiteLLM Pre-Auth SQL Injection Vulnerability (CVE-2026-42208)

A critical pre-authentication SQL injection vulnerability (CVE-2026-42208) in LiteLLM allows an unauthenticated remote attacker to extract all stored API keys and provider credentials, potentially leading to a complete compromise of connected AI provider accounts; active exploitation is reported, and the vulnerability is fixed in version v1.83.7.

CVE search metadata

CVE search record: CVE-2026-42208. Severity: critical. CVSS: 9.8. EPSS: 37.37%. KEV: yes. Product: LiteLLM. Brief: LiteLLM Pre-Auth SQL Injection Vulnerability (CVE-2026-42208). Brief link: https://feed.craftedsignal.io/briefs/2026-04-litellm-sqli/

Triage facts

Revision 1
Applicability
LiteLLM
Evidence
Confirmed activeHigh confidence
Severity
Critical
Sources
5 public sources
Freshness
Revision 1Apr 29, 15:21 UTC
Detection
2 rules
Vulnerabilities
1 CVE

What's new

  • 1. exploited:true; KEV CVE-2026-42208 Nov 20, 01:36 via cisa-kev
  • 2. poc_available May 10, 03:02 via sploitus
  • 3. added CVE-2026-42208 May 8, 17:25 via cisa-kev

Overview

LiteLLM is an AI gateway that centralizes API credentials for various providers like OpenAI and Anthropic. A pre-authentication SQL injection vulnerability, identified as CVE-2026-42208, allows an unauthenticated remote attacker to extract all stored API keys and provider credentials. Successful exploitation is equivalent to compromising all connected AI provider accounts. Open-source intelligence suggests active exploitation attempts in the wild. The vulnerability stems from directly concatenating the Bearer token value from incoming HTTP requests into a SQL query without proper sanitization before authentication. This allows for arbitrary SQL injection. The issue is resolved in LiteLLM version v1.83.7. Defenders should prioritize patching or mitigating this vulnerability immediately due to the high risk of credential compromise.

Attack Chain

  1. The attacker sends a malicious HTTP request to the LiteLLM server.
  2. The HTTP request contains a crafted Authorization: Bearer header.
  3. The Bearer token value contains a SQL injection payload.
  4. LiteLLM's authentication mechanism directly concatenates the Bearer token into a SQL query without sanitization.
  5. The injected SQL code executes within the LiteLLM database.
  6. The injected SQL is designed to extract API keys and provider credentials stored in the database.
  7. The extracted credentials are returned to the attacker within the HTTP response, or via an out-of-band channel configured in the injected SQL.
  8. The attacker uses the stolen API keys and credentials to access and control the victim's AI provider accounts.

Impact

Successful exploitation of this vulnerability allows an attacker to steal all API keys and credentials managed by the LiteLLM instance. This is equivalent to losing control over all connected AI provider accounts, including those for OpenAI and Anthropic. This may lead to unauthorized usage of AI services, data breaches, and other malicious activities. Open-source reporting indicates active exploitation, suggesting a high risk of widespread impact.

Recommendation

  • Immediately patch LiteLLM instances to version v1.83.7 to remediate CVE-2026-42208.
  • If patching is not immediately feasible, configure "set disable_error_logs: true" under general settings in LiteLLM to mitigate the vulnerability, as recommended by CCB.
  • Deploy the Sigma rule Detect Suspicious LiteLLM Authentication Header to identify potential exploitation attempts by monitoring for suspicious patterns in the Authorization header of HTTP requests to the LiteLLM server.
  • Upscale monitoring and detection capabilities, as recommended by CCB, to identify any related suspicious activity and ensure a swift response in case of an intrusion.

Detection coverage 2

Detect Suspicious LiteLLM Authentication Header

critical

Detects potential SQL injection attempts in the Authorization header of HTTP requests to LiteLLM, indicating exploitation of CVE-2026-42208

sigma tactics: initial_access techniques: T1190 sources: webserver, linux

Detect LiteLLM SQL Injection Error Logs

high

Detects SQL injection attempts based on specific error messages generated by LiteLLM, indicating a potential CVE-2026-42208 exploitation.

sigma tactics: initial_access techniques: T1190 sources: webserver, linux

Detection queries are available on the platform. Get full rules →