LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. From version 1.80.5 to before version 1.83.7, the POST /prompts/test endpoint accepted user-supplied prompt templates and rendered them without sandboxing. A crafted template could run arbitrary code inside the LiteLLM Proxy process. The endpoint only checks that the caller presents a valid proxy API key, so any authenticated user could reach it. Depending on how the proxy is deployed, this could expose secrets in the process environment (such as provider API keys or database credentials) and allow commands to be run on the host. This issue has been patched in version 1.83.7.
Weakness
The product uses a template engine to insert or process externally-influenced input, but it does not neutralize or incorrectly neutralizes special elements or syntax that can be interpreted as template expressions or other code directives when processed by the engine.
Affected Software
| Name | Vendor | Start Version | End Version |
|---|---|---|---|
| Litellm | Litellm | 1.80.5 (including) | 1.83.7 (excluding) |
Extended Description
Many web applications use template engines that allow developers to insert externally-influenced values into free text or messages in order to generate a full web page, document, message, etc. Such engines include Twig, Jinja2, Pug, Java Server Pages, FreeMarker, Velocity, ColdFusion, Smarty, and many others - including PHP itself. Some CMS (Content Management Systems) also use templates. Template engines often have their own custom command or expression language. If an attacker can influence input into a template before it is processed, then the attacker can invoke arbitrary expressions, i.e. perform injection attacks. For example, in some template languages, an attacker could inject the expression “{{7*7}}” and determine if the output returns “49” instead. The syntax varies depending on the language. In some cases, XSS-style attacks can work, which can obscure the root cause if the developer does not closely investigate the root cause of the error. Template engines can be used on the server or client, so both “sides” could be affected by injection. The mechanisms of attack or the affected technologies might be different, but the mistake is fundamentally the same.
Potential Mitigations
References
- https://github.com/BerriAI/litellm/releases/tag/v1.83.7-stable
- https://github.com/BerriAI/litellm/security/advisories/GHSA-xqmj-j6mv-4862
- https://access.redhat.com/security/cve/CVE-2026-42203
- https://bugzilla.redhat.com/show_bug.cgi?id=2467917
- https://security.access.redhat.com/data/csaf/v2/vex/2026/cve-2026-42203.json