Kubeflow

Enumerate and exploit Kubeflow Pipelines API instances.

Overview

The kubeflow module targets the Kubeflow Pipelines REST API (/pipeline/api/v1beta1/). It enumerates pipelines, runs, and experiments, lists Kubeflow Notebooks, and tests unauthenticated pipeline run creation. Kubeflow Pipelines is commonly deployed on shared ML platforms with misconfigured authentication, enabling direct API access to production ML workflows.

Subcommands

Read-Only (no --force-exploit required)

Subcommand	Description
enum	Probe API reachability and detect API version
pipelines	List accessible ML pipelines and their parameters
runs	List pipeline runs with status and pipeline correlation
experiments	List experiments
notebooks	List Kubeflow Notebooks in a namespace

Gated (requires --force-exploit)

Subcommand	Description
run-pipeline	Inject a new pipeline run via the Pipelines API

Flags

Flag	Required	Description
--target	Yes	Kubeflow URL (default port 8080)
--header	No	Custom HTTP headers. Repeatable.
--namespace	No	Kubernetes namespace for notebook listing (default: kubeflow)
--pipeline-id	For run-pipeline	Pipeline ID to execute
--experiment-id	No	Experiment ID for the new run
--run-name	For run-pipeline	Name for the new run
--param	No	Pipeline parameters as key=value pairs. Repeatable.

Key Endpoints

Endpoint	Method	Purpose
/pipeline/api/v1beta1/pipelines	GET	List pipelines (page_size=50, sort by created_at)
/pipeline/api/v1beta1/runs	GET	List runs
/pipeline/api/v1beta1/runs	POST	Create a new run (gated)
/pipeline/api/v1beta1/experiments	GET	List experiments
/notebook/api/namespaces/{ns}/notebooks	GET	List Kubeflow Notebooks
/pipeline/	GET	Dashboard fallback — used as reachability probe when v1beta1 is unavailable

Reachability Detection

The enum subcommand first probes GET /pipeline/api/v1beta1/pipelines?page_size=1. If that returns an error, it falls back to GET /pipeline/ to confirm dashboard reachability. A successful API probe sets APIVersion=v1beta1.

Examples

# Enumerate API reachability
aipostex kubeflow --target http://10.0.0.30:8080 enum

# List ML pipelines and parameters
aipostex kubeflow --target http://10.0.0.30:8080 pipelines

# List pipeline run history
aipostex kubeflow --target http://10.0.0.30:8080 runs

# List experiments
aipostex kubeflow --target http://10.0.0.30:8080 experiments

# List Kubeflow Notebooks in a namespace
aipostex kubeflow --target http://10.0.0.30:8080 notebooks --namespace kubeflow

# Inject a pipeline run (gated)
aipostex kubeflow --target http://10.0.0.30:8080 \
  run-pipeline --pipeline-id <pipeline-id> --run-name injected \
  --param learning_rate=0.1 --force-exploit

Workflow Progression

discover network (discovers Kubeflow on :8080)
  -> kubeflow enum (API version, reachability)
    -> kubeflow pipelines (pipeline inventory and parameters)
    -> kubeflow runs (run history and status)
    -> kubeflow experiments (experiment listing)
    -> kubeflow notebooks (notebook inventory)
    -> kubeflow run-pipeline (pipeline run injection, gated)

Vulnerability Templates

Template	Tags	Description
kubeflow-dashboard-unauth	kubeflow	Unauthenticated Kubeflow dashboard access
kubeflow-enum-001-pipeline-access	kubeflow, mlops, pipeline	Unauthenticated pipeline/experiment enumeration (read)