Glossary

AI Gateway

The central component that routes an application's LLM/AI calls: a single endpoint for multiple providers (OpenAI, Anthropic, self-hosted), with the keys, quotas, retries, and usage tracking. Examples: GitLab AI Gateway, Kong AI Gateway, LiteLLM. Since it concentrates **the providers' API keys** and **the content of the prompts**, an AI Gateway must be audited like an API gateway: least privilege on the config, log auditing, key rotation.

What to test on an AI Gateway

Who can read the config (provider keys), who can read the logs (other users' prompts), whether the endpoint requires authentication, whether a key can be injected via the request, and whether spend limits are set per user/team. The GitLab 18.6 → 18.8.1 case (CVE-2026-8810) shows the two classic leaks: a key readable by a Reporter, and logs containing other users' prompts.

Related Pages

Other Terms

Want to know what your AI agent can do?

Tell us about your agent, its tools, and client context. We will come back with the right review scope.

Discuss your audit