threatdeck
Live 14:45 UTC ·

AI Stack Alert High CVE-2026-116690 For SOC / IR, AppSec, AI Eng, CISO AI-assisted

MLflow 3.10.0–3.16.1: arbitrary file read via prompt model version source (CVE-2026-116690)

MLflow versions 3.10.0 through 3.16.1 fail to validate schemeless local paths in the prompt branch of CreateModelVersion. An attacker can store a path such as ../../../tmp as a model version source and read files from the server filesystem through the model-versions/get-artifact endpoint. The fix is in 3.17.0.

Why it matters

Unauthenticated mlflow server deployments, which are the default, expose readable server files to any network client that can reach the port. On basic-auth servers, a low-privilege user can read the authentication database, including the admin password hash.

Do this

  1. Upgrade MLflow to 3.17.0 or later on all tracking servers.
  2. If upgrade is not immediate, do not expose mlflow server to untrusted networks and place it behind authentication.
  3. Hunt server logs for model-versions/create requests with prompt tags and non-URI sources such as relative paths containing ../, and for get-artifact requests following them.
  4. Review the auth database and server file permissions, and rotate any admin credentials if exposure is suspected on basic-auth deployments.

Exploit Clock

Exploitation unlikely (for now)

  • EPSS 1.80%

Signals

  • NoCISA KEV
  • 7.5CVSS
  • 1.80%EPSS
  • 37/100Priority
  • 90%Confidence

Evidence

“a remote attacker can read arbitrary files from the server’s filesystem by creating a model version tagged as a prompt whose source is a schemeless local path”

GitHub Security Advisories (AI packages)

“Affected: MLflow 3.10.0 through 3.16.1. Fixed in 3.17.0 by”

GitHub Security Advisories (AI packages)

Sources

Affected / related

MLflow, mlflow, Path traversal

Published 2026-10-27 21:26 UTC · updated 2026-11-08 02:39 UTC · AI-assisted summary, checked against quoted sources · How we score

Following

Follow the vendors, products, AI tools, techniques and actors you care about. Matching stories, CVEs and exploited bugs are marked, a Following tab appears in the feed, and returning visits show what changed. Saved only in this browser.

Presets

Get updates

A personal RSS feed of new stories on everything you follow. Paste it into any feed reader, Slack or Teams RSS app.