CVE-2024-1741 is a critical (CVSS 9.1) improper authorization flaw in lunary-ai/lunary version 1.0.1. Removed members can read, create, modify and delete prompt templates by sending requests with a previously captured authorization token. Upgrade to a build with the fix and review your templates.
What's affected
Lunary is an open source platform for monitoring and managing applications built on large language models. The record references a GitHub fix commit and a huntr bounty report but does not name a first fixed release, so confirm the version in Lunary's release notes.
The record says that in version 1.0.1 members removed from an organization can still operate on prompt templates using their old authorization token. It describes the exposure as sensitive template data and the ability to change it.
The NVD score and vector are in the record: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:N. The weakness is CWE-863 (incorrect authorization).
How to check if you're exposed
If you self host Lunary, check the version you deployed and whether it includes the fix commit.
git merge-base --is-ancestor d8e2e73efd53ab4e92cf47bbf4b639a9f08853d2 HEAD && echo fixed
For container deployments, compare the image tag with the release notes. Review prompt templates for unexpected edits since people were removed.
What to do now
- Upgrade Lunary to a release that contains the fix commit.
- Review the membership of your organizations and projects, and remove accounts that should not have access.
- Rotate project API keys if a former member might have had access.
- Restrict who can sign up on a self hosted instance.
- Test authorization on every endpoint that takes an object ID.
Why it matters
Prompt templates shape what an application sends to a model, so unauthorized edits can change behavior without any code deploy. That makes this more than a data exposure. The pattern matches CVE-2024-1740, a sibling issue in the same version. The record does not flag exploitation. Safeguard's free in-browser dependency scan at gold.safeguard.sh/scan can show whether a component tied to this CVE appears in your dependency manifests. The Gold directory keeps the record, severity and references in one place, and reachability analysis helps you separate code paths that can actually be hit from packages that are merely installed.
A practical way to close this out is to record the affected component, the owner, the target version and a due date in your tracker, then verify the result after the change rather than assuming the update took. Re-run your dependency or inventory check a week later to catch machines that missed it.
Quick check
Is this the same as CVE-2024-1740?
No, but it is the same kind of flaw. One covers logs and the other covers prompt templates, both in version 1.0.1.
Sources
Last reviewed: 19 October 2026.
