This vulnerability in BerriAI litellm (versions 1.82.0 through 1.82.5) affects the async_pre_call_hook function in the enterprise_hooks/banned_keywords.py file of the Completions Interface component. The issue is caused by improper authorization checks when processing the prompt argument, which can be manipulated by an attacker remotely to bypass intended access controls. The vulnerability has a CVSS 4.0 base score of 5.3 (medium severity) with network attack vector, low complexity, no privileges required, and no user interaction needed. Although the vendor was contacted early, there is no available information on an official patch or remediation. Public exploit code is available, increasing the risk of exploitation.

An attacker can remotely exploit this vulnerability by manipulating the prompt argument to bypass authorization controls in the affected function. This could lead to unauthorized access or actions within the Completions Interface component of BerriAI litellm. The CVSS score indicates a medium impact with potential for partial confidentiality, integrity, and availability loss. Since exploit code is publicly available, the risk of exploitation is heightened.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is released, users should consider restricting network access to the affected component and monitor for suspicious activity related to prompt manipulation. Follow vendor communications closely for updates on patches or official mitigations.