--- /dev/null
+# Policy on AI Agents
+
+> [!IMPORTANT]
+> This project does **not** accept pull requests that are fully or
+> predominantly AI-generated. AI tools may be utilized solely to
+> assist individual development.
+>
+> If we receive an AI PR that _clearly has had zero testing_, then we
+> will **ban the person who submitted the PR**.
+>
+> Submitting a garbage PR isn't just AI slop, it's abuse.
+
+AI assistance is fine. But when someone submits a PR, we assume that
+they understand the submission, and that they accept full
+responsibility for it.
+
+---
+
+## Recommendations for Using AI
+
+AI can be useful in some situations. However, our experience has been
+that when people ask AI questions about FreeRADIUS, the answers are
+pretty bad. The advice is either unclear, or is simply wrong. This
+means that AI should be used to point you in a direction, but you
+should read the documentation to get the _truthful_ answers:
+
+> READ THE DOCS!
+>
+> https://www.freeradius.org/documentation/freeradius-server/
+
+AI can help in the following use cases:
+
+- Code completion in an editor,
+- Asking about the structure of the code,
+- Learning about specific techniques used in the project,
+- Finding documents, links, and parts of the code that are worth your time
+- Reviewing human-written code and providing suggestions on things you hadn't thought of,
+- Creating rough outlines for documentation,
+
+Anything AI generated _must_ be reviewed, edited, and approved by a person.
+
+## Disclosure
+
+**All AI contributions must be disclosed.**
+
+If you use AI in an editor for API suggestions, searching the code,
+etc., that doesn't need to be disclosed. You are the person writing
+the code, not the AI.
+
+If you use AI to _generate_ code, it **must** be disclosed. The code
+that AI generates is pretty much always wrong. We need to know if the
+PR is AI slop, or if it has been reviewed and understood by a person.