Exposure of Sensitive Information in berriai/litellm

Vulnerability

Overview

In LiteLLM version 1.52.1, the proxy_server.py file contains an error-handling flaw that leaks Langfuse API keys when a parsing error occurs while processing team settings [2]. The error messages inadvertently include the langfuse_secret and langfuse_public_key, exposing credentials meant to be confidential.

Exploitation

An attacker can exploit this by causing a parsing error in team settings, which could be triggered by sending a specially crafted request to the proxy server. No authentication is explicitly required beyond network access to the server, making this a low-barrier attack [2][3].

Impact

With the leaked Langfuse API keys, an attacker gains full access to the associated Langfuse project, which stores all requests processed through the proxy. This can lead to exposure of sensitive data, including prompts and responses from LLM interactions [2].

Mitigation

As of March 2025, the vulnerability exists in version 1.52.1. The issue has been reported via a bug bounty program [3]. Users should upgrade to a patched version once available, or apply workarounds such as restricting network access and monitoring error logs for exposed credentials.