CVE-2026-31020

DocsGPT 0.15.0 and below contains a server-side template injection (CWE-94) in its custom prompt feature, which renders user-supplied prompt content with Jinja templates without input sanitization or sandboxing. An unauthenticated attacker can submit prompt data containing malicious Jinja template expressions, which the server evaluates during chatbot interactions. Successful injection is stated to allow full remote code execution on the server running DocsGPT. Any deployment of DocsGPT 0.15.0 or earlier that exposes the chat/prompt interface to untrusted users is affected. Exploitation has not been reported: there is no known public proof of concept, the flaw is not in CISA's KEV catalog, and EPSS currently estimates a 0.6% chance of exploitation within 30 days.

What to do: Upgrade DocsGPT to a release newer than 0.15.0 as soon as a patched version is published, and check the project's advisories/releases for the fixed version. Until then, limit network exposure of DocsGPT instances (restrict to trusted networks or require authentication at a reverse proxy), and consider disabling or tightly restricting the custom prompt feature. Do not render untrusted user prompt content through unsandboxed Jinja templates in any customizations or integrations.