AI analysis
Claude Opus · 8 Nov 2026, 10:21 CET
LMFDB template injection lets unauthenticated users run server commands
LMFDB deployments built from code before commit 01829bd let unauthenticated attackers inject Jinja template expressions through the knowledge render endpoint and run operating system commands. Update to code that includes commit 01829bd, or block the endpoint until you can.
In plain terms
LMFDB is a web application built with Python. One of its features turns stored text snippets, called knowls, into web pages using a template engine called Jinja. A template engine fills placeholders in a page with data. In affected versions, text supplied by a visitor in the web address or in submitted content was treated as part of the template itself rather than as plain data. Anyone on the internet could therefore send instructions that the server would carry out, without logging in.
This matters to anyone who runs their own LMFDB instance, such as a mirror, a development server or a research deployment. An attacker could run commands on the server with the same permissions as the web application. That could expose data, alter the site or disrupt it.