Security Scan Finds Nearly 10% of Public LiteLLM Gateways Exposed to Default Admin Key

A security analysis conducted by Wiz Research in February revealed that close to one in ten internet-facing LiteLLM gateways accepted the placeholder admin key "sk-1234"—the example credential shown in the project's own setup documentation. The finding highlights a widespread configuration oversight among deployments of the open‑source AI gateway, which many organizations use to route requests from their applications to large language models.

LiteLLM, a community‑maintained project, serves as a middleware layer that abstracts model providers, enforces usage policies, and can manage billing. Because the software is freely available, it is often deployed by developers with limited security expertise. The research team scanned hundreds of publicly reachable LiteLLM instances and identified that the default key, intended only for illustration, remained active on a significant subset of servers.

The presence of an active example key effectively grants unrestricted administrative access, allowing an attacker to alter routing rules, harvest usage data, or even execute arbitrary code depending on the gateway's configuration. While the scan did not confirm any malicious exploitation, the vulnerability aligns with a broader pattern of default‑credential exposures that have plagued other open‑source tools, from database servers to content management systems.

Following the disclosure, the LiteLLM maintainers issued an advisory urging users to replace the sample key with a strong, unique secret and to audit their deployments for similar misconfigurations. They also updated the project's documentation to emphasize the security implications of leaving the example credential unchanged. Security experts recommend that organizations treating LiteLLM as a production component adopt standard hardening practices, such as network segmentation, rate limiting, and regular credential rotation.

Analysts say the incident underscores the importance of secure default settings in open‑source software, especially as AI services become more integral to business workflows. As the ecosystem matures, developers and operators are expected to adopt stricter security controls, and the community around LiteLLM is likely to prioritize automated checks that flag default credentials during installation. Continued monitoring by security researchers will be essential to ensure that similar oversights are identified and remediated before they can be leveraged in real‑world attacks.