LiteLLM's proxy layer turns into a key vault breach when this flaw is hit. A pre-auth SQL injection in the API key verification path lets unauthenticated attackers read and modify the database that holds master keys, API keys, and other secrets. Patching closes the injection point, but it does not undo any credentials already exposed.
The issue is CVE-2026-42208. LiteLLM fixed it in version 1.83.7 by replacing string concatenation with parameterized queries. Reporting and vendor guidance say exploitation began about 36 hours after disclosure, and attackers were targeting tables with provider credentials, environment data, and configs rather than probing randomly.
For teams using LiteLLM as a shared gateway, the real risk is impersonation of downstream services after database access. Once those keys are taken, attackers can abuse connected LLM instances and other cloud services even if the gateway itself is later patched.