PT-2026-127745

تاريخ النشر

2026-10-27

·

تاريخ التحديث

2026-10-27

الوصف من

CVSS 3.1
9.8
حرجة
المقاييسAV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H

Summary

MLflow 3.12.0 introduced registration and deserialization support for third-party scorers in OSS MLflow. During scorer deserialization, the server accepts a request-controlled Python module path beneath one of four allowed package prefixes, imports that module, obtains a request-controlled class, and instantiates it synchronously.
An HTTP caller can combine this behavior with MLflow's built-in experiment, run, and artifact-upload endpoints:
  1. Create an experiment whose local artifact location points to a writable allowed scorer package.
  2. Create a run, causing MLflow to allocate <artifact location>/<run id>/artifacts/.
  3. Upload a Python module into that run artifact directory.
  4. Submit a serialized third-party scorer whose module path resolves to the uploaded file.
  5. Trigger server-side import and class construction through /ajax-api/3.0/mlflow/scorer/invoke.
On the default no-auth server using the official MLflow v3.14.0 image, this produces unauthenticated arbitrary Python code execution with the privileges of the MLflow server process. The reproduced image runs the process as uid 0 and permits writes to its installed Python package tree.
The attached proof of concept is intentionally harmless. The uploaded constructor performs one action only: it writes a random marker beside its own uploaded run artifact. The marker is then downloaded through MLflow's run-artifact endpoint and checked for an exact match.

Technical details

1. The allow-list accepts attacker-created descendants

At release commit 86cd7f5a1dc25a1387ad07c87811bbc30f62951c, mlflow/genai/scorers/scorer utils.py defines these allowed roots:
python
THIRD PARTY SCORER ALLOWED MODULES = frozenset({
  "mlflow.genai.scorers.ragas",
  "mlflow.genai.scorers.deepeval",
  "mlflow.genai.scorers.trulens",
  "mlflow.genai.scorers.phoenix",
})
In mlflow/genai/scorers/base.py, the deserializer accepts either an exact root or any descendant:
python
module path == m or module path.startswith(m + ".")
It then performs the following operations using request-controlled values:
python
module = importlib.import module(module path)
scorer class = getattr(module, class name, None)
scorer instance = scorer class(**init kwargs)
A package-name prefix is not a provenance boundary. If the server process can write below an allowed package, an attacker-created descendant can satisfy the string-prefix check and become importable Python code.
Relevant source locations at the v3.14.0 release commit:
  • mlflow/genai/scorers/scorer utils.py, allowed third-party scorer modules
  • mlflow/genai/scorers/base.py, third-party scorer deserialization and construction

2. Deserialization occurs before trace processing

In mlflow/server/handlers.py, invoke scorer handler() parses the supplied serialized scorer and calls:
python
scorer = Scorer.model validate json(serialized scorer)
Only after deserialization and construction does the handler call:
python
batches = get trace batches for scorer(trace ids, scorer, tracking store)
The attacker-controlled constructor therefore runs before the server validates or resolves the supplied trace IDs.
The minimal proof class is deliberately not a complete MLflow scorer. After its constructor writes the marker, later trace-batching logic accesses scorer attributes that the proof class does not implement. The resulting HTTP 500 occurs after code execution and does not invalidate the proof.
No background job execution or special job-execution server option is required to reach the constructor.

3. MLflow's artifact layout places the module under the trusted prefix

The experiment creation endpoint accepts an artifact location. For file-backed experiments, MLflow derives each run artifact directory as:
text
<artifact location>/<run id>/artifacts/
The proof of concept selects this package directory in the official v3.14.0 image:
text
file:///usr/local/lib/python3.10/site-packages/mlflow/genai/scorers/trulens
After creating a run and uploading audit <token>.py, the resulting filesystem location is:
text
/usr/local/lib/python3.10/site-packages/mlflow/genai/scorers/trulens/<run id>/artifacts/audit <token>.py
The corresponding import path is:
text
mlflow.genai.scorers.trulens.<run id>.artifacts.audit <token>
The attack does not overwrite an existing package module. It creates a new run-scoped module in the normal file-backed artifact layout. The run ID is returned to the caller.

4. Authentication and authorization behavior

With authentication disabled, as in the default reproduced server configuration, no credentials are required.
When MLflow basic authentication is enabled, authentication still applies. However, the scorer-invocation AJAX endpoint is assigned to validate gateway proxy, which returns True without performing an experiment- or scorer-level resource permission check.
Practical exploitation in an authenticated deployment still requires the caller to create an experiment and run and upload artifacts to them. The key authorization issue is that scorer deserialization itself is not gated by a resource-specific permission check for the supplied experiment before the import occurs.

Proof of concept

The attached reproduction is intentionally isolated:
  • exact official MLflow v3.14.0 image pinned by manifest digest;
  • Docker network disabled;
  • MLflow bound to container loopback;
  • no shell command or external callback;
  • no external request;
  • no host filesystem write;
  • marker created beside the uploaded run artifact;
  • marker retrieved through the run-artifact download endpoint; and
  • disposable container removed after completion.
Run:
bash
./run-official-image.sh
Expected output:
text
MLFLOW VERSION=3.14.0
PACKAGE DIR=/usr/local/lib/python3.10/site-packages/mlflow/genai/scorers/trulens
CREATE EXPERIMENT STATUS=200
CREATE RUN STATUS=200
UPLOAD STATUS=200
INVOKE STATUS=500
PROOF DOWNLOAD STATUS=200
PROOF MATCH=1
RESULT=AFFECTED
The security-relevant result is:
text
PROOF MATCH=1
RESULT=AFFECTED
This proves that the attacker-controlled constructor executed in the MLflow server process before the later scorer-processing failure.

Reproduction environment

  • Release: MLflow v3.14.0
  • Release commit: 86cd7f5a1dc25a1387ad07c87811bbc30f62951c
  • Official image: ghcr.io/mlflow/mlflow:v3.14.0
  • Manifest digest: sha256:967de3ca34b6a5751718e380a0a383eb0ce1a6f5b484f363bd0da7559117a5f8
  • Python version: 3.10.18
  • Image-configured user: empty
  • Observed runtime user: uid 0
  • Server bind: 127.0.0.1:5000
  • Container network: disabled
  • Backend store: sqlite:////work/mlflow.db
  • Default artifact root: /work/artifacts
  • Authentication: disabled
  • Special job-execution option: none
Server command inside the disposable container:
bash
mlflow server 
 --host 127.0.0.1 
 --port 5000 
 --backend-store-uri sqlite:////work/mlflow.db 
 --default-artifact-root /work/artifacts

Affected range and source lineage

The vulnerable third-party scorer deserialization behavior was introduced by commit:
text
02eea26035b9d8e65af31e6e8f058ca195f62cdd
Enable third-party scorer registration in OSS MLflow (#22634)
The feature is absent from v3.11.1 and first appears in v3.12.0.
Runtime reproduction succeeds on v3.14.0. The relevant mlflow/genai/scorers/base.py content is unchanged at master commit:
text
20858c9426a9644dfcf9545a92cbb534376779eb
No fixed release was identified as of 2026-07-12.

Impact

On the reproduced stock no-auth image, any network caller who can reach the tracking server can execute Python code as the MLflow process without credentials or user interaction.
Successful exploitation provides confidentiality, integrity, and availability impact over all data, credentials, network access, and filesystem locations available to the MLflow server process.
The stock image's runtime uid 0 increases container-level impact. Host impact still depends on the container runtime configuration, mounted paths, granted capabilities, socket exposure, and other isolation controls.
In authentication-enabled deployments, exploitation requires an authenticated principal capable of creating an experiment and run and uploading artifacts. The exact chain also requires the server process to be able to write beneath one of the allowed scorer packages. This condition holds in the reproduced official image and may hold in virtual environments installed and run by the same operating-system user.
A read-only container filesystem or a scorer package directory owned by a different, non-writable user blocks this exact placement chain.

Artifacts and attachments

Remote proof-of-concept script

POC Script: [new remote poc.py](https://github.com/user-attachments/files/29936616/new remote poc.py)
The remote proof of concept was then executed as follows. The disposable machine’s public IP has been redacted:
bash
python3 new remote poc.py 
 --target http://<REDACTED PUBLIC IP> 
 --package-dir /usr/local/lib/python3.10/site-packages/mlflow/genai/scorers/trulens 
 --acknowledge-owned-target I OWN THIS TARGET
The complete sanitized output was:
text
TARGET=http://<REDACTED PUBLIC IP>
VERSION STATUS=200
MLFLOW VERSION=3.14.0
CREATE EXPERIMENT STATUS=200
EXPERIMENT ID=2
CREATE RUN STATUS=200
RUN ID=9a2d665eb38940b68983f0586f623115
UPLOAD STATUS=200
IMPORT PATH=mlflow.genai.scorers.trulens.9a2d665eb38940b68983f0586f623115.artifacts.audit 1783843760683972103
INVOKE STATUS=500
INVOKE RESPONSE PREVIEW='<!doctype html>
<html lang=en>
<title>500 Internal Server Error</title>
<h1>Internal Server Error</h1>
<p>The server encountered an internal error and was unable to complete your request. Either the server is overloaded or there is an error in the application.</p>
'
PROOF DOWNLOAD STATUS=200
EXPECTED MARKER SHA256=7a423b479df3524567fdf8cc3e7f6c4bfccc577fbcc4bac9519db8bd1cc61ed5
PROOF MATCH=1
SERVER CLEANUP PATH=/usr/local/lib/python3.10/site-packages/mlflow/genai/scorers/trulens/9a2d665eb38940b68983f0586f623115
PROOF OUTPUT BEGIN
{
 "cwd": "/",
 "machine": "x86 64",
 "marker": "MLFLOW REMOTE SAFE MARKER 1783843760683972103",
 "os release": "PRETTY NAME="Debian GNU/Linux 11 (bullseye)"
NAME="Debian GNU/Linux"
VERSION ID="11"
VERSION="11 (bullseye)"
VERSION CODENAME=bullseye
ID=debian
HOME URL="https://www.debian.org/"
SUPPORT URL="https://www.debian.org/support"
BUG REPORT URL="https://bugs.debian.org/"
",
 "platform": "Linux-7.0.0-1006-aws-x86 64-with-glibc2.31",
 "python implementation": "CPython",
 "python version": "3.10.18",
 "server datetime utc": "2026-07-12T08:09:43.555055+00:00",
 "whoami": "root",
 "whoami returncode": 0,
 "whoami stderr": ""
}
PROOF OUTPUT END
RESULT=AFFECTED
EXPERIMENT METADATA CLEANUP STATUS=200
CLEANUP NOTE=discard the temporary test instance or remove SERVER CLEANUP PATH on the owned host
The proof file was also verified directly inside the disposable container:
bash
cat /usr/local/lib/python3.10/site-packages/mlflow/genai/scorers/trulens/9a2d665eb38940b68983f0586f623115/artifacts/execution-proof-1783843760683972103.json
The file contained:
json
{
 "cwd": "/",
 "machine": "x86 64",
 "marker": "MLFLOW REMOTE SAFE MARKER 1783843760683972103",
 "os release": "PRETTY NAME="Debian GNU/Linux 11 (bullseye)"
NAME="Debian GNU/Linux"
VERSION ID="11"
VERSION="11 (bullseye)"
VERSION CODENAME=bullseye
ID=debian
HOME URL="https://www.debian.org/"
SUPPORT URL="https://www.debian.org/support"
BUG REPORT URL="https://bugs.debian.org/"
",
 "platform": "Linux-7.0.0-1006-aws-x86 64-with-glibc2.31",
 "python implementation": "CPython",
 "python version": "3.10.18",
 "server datetime utc": "2026-07-12T08:09:43.555055+00:00",
 "whoami": "root",
 "whoami returncode": 0,
 "whoami stderr": ""
}
The results demonstrate that:
  • the attacker-controlled module was uploaded beneath an allowed scorer package;
  • MLflow imported the request-selected module and invoked its constructor;
  • the constructor executed as the MLflow server process, which ran as root in the reproduced official image;
  • the constructor created a server-side proof file containing a random marker and bounded runtime information;
  • the proof file was retrieved successfully through MLflow’s artifact-download endpoint; and
  • the later HTTP 500 occurred only after constructor execution.
The authoritative reproduction indicators are:
text
UPLOAD STATUS=200
PROOF DOWNLOAD STATUS=200
PROOF MATCH=1
RESULT=AFFECTED
The disposable cloud machine and container were removed after testing.
Local Reproduction Scripts: run-official-image.sh [poc local.py](https://github.com/user-attachments/files/29936506/poc local.py)
قابلية الاستغلال
إصلاح
التأثيرات
المورّدون
المنتجات المتأثرة
تصنيف نقاط الضعف
CWE-470: Use of Externally-Controlled Input to Select Classes or Code (\'Unsafe Reflection\')
CWE-94: Improper Control of Generation of Code (\'Code Injection\')
المعرّفات ذات الصلة
GHSA-26P8-2JQ9-3VQ9
وجدت خطأً في الوصف؟ لديك ما تضيفه؟ لا تتردد في مراسلتنا 👾
dbugs@ptsecurity.com