The vulnerability resides in the Admin Key Handler component of BerriAI litellm, specifically within an undisclosed function in litellm/proxy/management_endpoints/key_management_endpoints.py. It allows improper authorization of requests to key management endpoints, meaning an attacker can access, create, or delete administrative keys without proper authentication. This flaw can lead to unauthorized use of language‑model API keys, potentially enabling malicious usage or secret exfiltration.

All installations of BerriAI litellm up to and including version 1.63.1 are affected. The product is identified as litellm from BerriAI and the associated CPE confirms the same. No later versions are known to contain the fix.

The CVSS score of 5.3 indicates a moderate severity, and the EPSS score is not available, but the lack of public exploitation yet does not preclude the risk. The vulnerability can be triggered remotely by any user able to reach the key management endpoints, and a publicly disclosed exploit is already available. The vendor has issued a patch through pull request 23781; however, until applied the flaw remains exploitable.