{
  "affected": [
    {
      "affectedData": [
        {
          "product": "mlflow",
          "vendor": "mlflow",
          "versions": [
            {
              "status": "affected",
              "version": "\u003e= 3.12.0rc0, \u003c 3.17.0"
            }
          ]
        }
      ],
      "source": "security-advisories@github.com"
    }
  ],
  "cveTags": [],
  "descriptions": [
    {
      "lang": "en",
      "value": "MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. From 3.12.0rc0 until 3.17.0, third-party scorer deserialization in mlflow/genai/scorers/base.py accepts request-selected descendants of THIRD_PARTY_SCORER_ALLOWED_MODULES rather than requiring an exact allowed module. A caller can use MLflow artifact endpoints to place a Python module beneath a writable allowed package and submit its module and class through /ajax-api/3.0/mlflow/scorer/invoke. Scorer.model_validate_json() imports the module and constructs the class before trace processing. Authentication-disabled deployments permit this code execution without credentials, while basic-auth deployments require a principal capable of creating an experiment and run and uploading artifacts. In either configuration, the allowed scorer package must be writable by the server process. This issue is fixed in version 3.17.0."
    }
  ],
  "id": "CVE-2026-116692",
  "lastModified": "2026-10-27T21:18:05.013",
  "metrics": {
    "cvssMetricV31": [
      {
        "cvssData": {
          "attackComplexity": "LOW",
          "attackVector": "NETWORK",
          "availabilityImpact": "HIGH",
          "baseScore": 9.8,
          "baseSeverity": "CRITICAL",
          "confidentialityImpact": "HIGH",
          "integrityImpact": "HIGH",
          "privilegesRequired": "NONE",
          "scope": "UNCHANGED",
          "userInteraction": "NONE",
          "vectorString": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H",
          "version": "3.1"
        },
        "exploitabilityScore": 3.9,
        "impactScore": 5.9,
        "source": "security-advisories@github.com",
        "type": "Secondary"
      }
    ]
  },
  "published": "2026-10-27T21:18:05.013",
  "references": [
    {
      "source": "security-advisories@github.com",
      "url": "https://github.com/mlflow/mlflow/commit/a732443e695c3c677680c54320efc394c81654ab"
    },
    {
      "source": "security-advisories@github.com",
      "url": "https://github.com/mlflow/mlflow/pull/25906"
    },
    {
      "source": "security-advisories@github.com",
      "url": "https://github.com/mlflow/mlflow/releases/tag/v3.17.0"
    },
    {
      "source": "security-advisories@github.com",
      "url": "https://github.com/mlflow/mlflow/security/advisories/GHSA-26p8-2jq9-3vq9"
    }
  ],
  "sourceIdentifier": "security-advisories@github.com",
  "vulnStatus": "Received",
  "weaknesses": [
    {
      "description": [
        {
          "lang": "en",
          "value": "CWE-94"
        },
        {
          "lang": "en",
          "value": "CWE-470"
        }
      ],
      "source": "security-advisories@github.com",
      "type": "Primary"
    }
  ]
}

{
  "affected": [
    {
      "package": {
        "ecosystem": "PyPI",
        "name": "mlflow"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "3.12.0rc0"
            },
            {
              "fixed": "3.17.0"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-116692"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-470",
      "CWE-94"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-10-27T21:26:32Z",
    "nvd_published_at": null,
    "severity": "CRITICAL"
  },
  "details": "## Summary\n\nMLflow 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.\n\nAn HTTP caller can combine this behavior with MLflow\u0027s built-in experiment, run, and artifact-upload endpoints:\n\n1. Create an experiment whose local artifact location points to a writable allowed scorer package.\n2. Create a run, causing MLflow to allocate `\u003cartifact_location\u003e/\u003crun_id\u003e/artifacts/`.\n3. Upload a Python module into that run artifact directory.\n4. Submit a serialized third-party scorer whose module path resolves to the uploaded file.\n5. Trigger server-side import and class construction through `/ajax-api/3.0/mlflow/scorer/invoke`.\n\nOn 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.\n\nThe 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\u0027s run-artifact endpoint and checked for an exact match.\n\n## Technical details\n\n### 1. The allow-list accepts attacker-created descendants\n\nAt release commit `86cd7f5a1dc25a1387ad07c87811bbc30f62951c`, `mlflow/genai/scorers/scorer_utils.py` defines these allowed roots:\n\n```python\nTHIRD_PARTY_SCORER_ALLOWED_MODULES = frozenset({\n    \"mlflow.genai.scorers.ragas\",\n    \"mlflow.genai.scorers.deepeval\",\n    \"mlflow.genai.scorers.trulens\",\n    \"mlflow.genai.scorers.phoenix\",\n})\n```\n\nIn `mlflow/genai/scorers/base.py`, the deserializer accepts either an exact root or any descendant:\n\n```python\nmodule_path == m or module_path.startswith(m + \".\")\n```\n\nIt then performs the following operations using request-controlled values:\n\n```python\nmodule = importlib.import_module(module_path)\nscorer_class = getattr(module, class_name, None)\nscorer_instance = scorer_class(**init_kwargs)\n```\n\nA 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.\n\nRelevant source locations at the v3.14.0 release commit:\n\n* `mlflow/genai/scorers/scorer_utils.py`, allowed third-party scorer modules\n* `mlflow/genai/scorers/base.py`, third-party scorer deserialization and construction\n\n### 2. Deserialization occurs before trace processing\n\nIn `mlflow/server/handlers.py`, `_invoke_scorer_handler()` parses the supplied serialized scorer and calls:\n\n```python\nscorer = Scorer.model_validate_json(serialized_scorer)\n```\n\nOnly after deserialization and construction does the handler call:\n\n```python\nbatches = get_trace_batches_for_scorer(trace_ids, scorer, tracking_store)\n```\n\nThe attacker-controlled constructor therefore runs before the server validates or resolves the supplied trace IDs.\n\nThe 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.\n\nNo background job execution or special job-execution server option is required to reach the constructor.\n\n### 3. MLflow\u0027s artifact layout places the module under the trusted prefix\n\nThe experiment creation endpoint accepts an `artifact_location`. For file-backed experiments, MLflow derives each run artifact directory as:\n\n```text\n\u003cartifact_location\u003e/\u003crun_id\u003e/artifacts/\n```\n\nThe proof of concept selects this package directory in the official v3.14.0 image:\n\n```text\nfile:///usr/local/lib/python3.10/site-packages/mlflow/genai/scorers/trulens\n```\n\nAfter creating a run and uploading `audit_\u003ctoken\u003e.py`, the resulting filesystem location is:\n\n```text\n/usr/local/lib/python3.10/site-packages/mlflow/genai/scorers/trulens/\u003crun_id\u003e/artifacts/audit_\u003ctoken\u003e.py\n```\n\nThe corresponding import path is:\n\n```text\nmlflow.genai.scorers.trulens.\u003crun_id\u003e.artifacts.audit_\u003ctoken\u003e\n```\n\nThe 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.\n\n### 4. Authentication and authorization behavior\n\nWith authentication disabled, as in the default reproduced server configuration, no credentials are required.\n\nWhen 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.\n\nPractical 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.\n\n## Proof of concept\n\nThe attached reproduction is intentionally isolated:\n\n* exact official MLflow v3.14.0 image pinned by manifest digest;\n* Docker network disabled;\n* MLflow bound to container loopback;\n* no shell command or external callback;\n* no external request;\n* no host filesystem write;\n* marker created beside the uploaded run artifact;\n* marker retrieved through the run-artifact download endpoint; and\n* disposable container removed after completion.\n\nRun:\n\n```bash\n./run-official-image.sh\n```\n\nExpected output:\n\n```text\nMLFLOW_VERSION=3.14.0\nPACKAGE_DIR=/usr/local/lib/python3.10/site-packages/mlflow/genai/scorers/trulens\nCREATE_EXPERIMENT_STATUS=200\nCREATE_RUN_STATUS=200\nUPLOAD_STATUS=200\nINVOKE_STATUS=500\nPROOF_DOWNLOAD_STATUS=200\nPROOF_MATCH=1\nRESULT=AFFECTED\n```\n\nThe security-relevant result is:\n\n```text\nPROOF_MATCH=1\nRESULT=AFFECTED\n```\n\nThis proves that the attacker-controlled constructor executed in the MLflow server process before the later scorer-processing failure.\n\n## Reproduction environment\n\n* **Release:** MLflow v3.14.0\n* **Release commit:** `86cd7f5a1dc25a1387ad07c87811bbc30f62951c`\n* **Official image:** `ghcr.io/mlflow/mlflow:v3.14.0`\n* **Manifest digest:** `sha256:967de3ca34b6a5751718e380a0a383eb0ce1a6f5b484f363bd0da7559117a5f8`\n* **Python version:** 3.10.18\n* **Image-configured user:** empty\n* **Observed runtime user:** uid 0\n* **Server bind:** `127.0.0.1:5000`\n* **Container network:** disabled\n* **Backend store:** `sqlite:////work/mlflow.db`\n* **Default artifact root:** `/work/artifacts`\n* **Authentication:** disabled\n* **Special job-execution option:** none\n\nServer command inside the disposable container:\n\n```bash\nmlflow server \\\n  --host 127.0.0.1 \\\n  --port 5000 \\\n  --backend-store-uri sqlite:////work/mlflow.db \\\n  --default-artifact-root /work/artifacts\n```\n\n## Affected range and source lineage\n\nThe vulnerable third-party scorer deserialization behavior was introduced by commit:\n\n```text\n02eea26035b9d8e65af31e6e8f058ca195f62cdd\nEnable third-party scorer registration in OSS MLflow (#22634)\n```\n\nThe feature is absent from v3.11.1 and first appears in v3.12.0.\n\nRuntime reproduction succeeds on v3.14.0. The relevant `mlflow/genai/scorers/base.py` content is unchanged at master commit:\n\n```text\n20858c9426a9644dfcf9545a92cbb534376779eb\n```\n\nNo fixed release was identified as of 2026-07-12.\n\n## Impact\n\nOn 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.\n\nSuccessful exploitation provides confidentiality, integrity, and availability impact over all data, credentials, network access, and filesystem locations available to the MLflow server process.\n\nThe stock image\u0027s 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.\n\nIn 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.\n\nA read-only container filesystem or a scorer package directory owned by a different, non-writable user blocks this exact placement chain.\n\n## Artifacts and attachments\n\n### Remote proof-of-concept script\n\nPOC Script: [new_remote_poc.py](https://github.com/user-attachments/files/29936616/new_remote_poc.py)\n\n\nThe remote proof of concept was then executed as follows. The disposable machine\u2019s public IP has been redacted:\n\n```bash\npython3 new_remote_poc.py \\\n  --target http://\u003cREDACTED_PUBLIC_IP\u003e \\\n  --package-dir /usr/local/lib/python3.10/site-packages/mlflow/genai/scorers/trulens \\\n  --acknowledge-owned-target I_OWN_THIS_TARGET\n```\n\nThe complete sanitized output was:\n\n```text\nTARGET=http://\u003cREDACTED_PUBLIC_IP\u003e\nVERSION_STATUS=200\nMLFLOW_VERSION=3.14.0\nCREATE_EXPERIMENT_STATUS=200\nEXPERIMENT_ID=2\nCREATE_RUN_STATUS=200\nRUN_ID=9a2d665eb38940b68983f0586f623115\nUPLOAD_STATUS=200\nIMPORT_PATH=mlflow.genai.scorers.trulens.9a2d665eb38940b68983f0586f623115.artifacts.audit_1783843760683972103\nINVOKE_STATUS=500\nINVOKE_RESPONSE_PREVIEW=\u0027\u003c!doctype html\u003e\\n\u003chtml lang=en\u003e\\n\u003ctitle\u003e500 Internal Server Error\u003c/title\u003e\\n\u003ch1\u003eInternal Server Error\u003c/h1\u003e\\n\u003cp\u003eThe 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.\u003c/p\u003e\\n\u0027\nPROOF_DOWNLOAD_STATUS=200\nEXPECTED_MARKER_SHA256=7a423b479df3524567fdf8cc3e7f6c4bfccc577fbcc4bac9519db8bd1cc61ed5\nPROOF_MATCH=1\nSERVER_CLEANUP_PATH=/usr/local/lib/python3.10/site-packages/mlflow/genai/scorers/trulens/9a2d665eb38940b68983f0586f623115\nPROOF_OUTPUT_BEGIN\n{\n  \"cwd\": \"/\",\n  \"machine\": \"x86_64\",\n  \"marker\": \"MLFLOW_REMOTE_SAFE_MARKER_1783843760683972103\",\n  \"os_release\": \"PRETTY_NAME=\\\"Debian GNU/Linux 11 (bullseye)\\\"\\nNAME=\\\"Debian GNU/Linux\\\"\\nVERSION_ID=\\\"11\\\"\\nVERSION=\\\"11 (bullseye)\\\"\\nVERSION_CODENAME=bullseye\\nID=debian\\nHOME_URL=\\\"https://www.debian.org/\\\"\\nSUPPORT_URL=\\\"https://www.debian.org/support\\\"\\nBUG_REPORT_URL=\\\"https://bugs.debian.org/\\\"\\n\",\n  \"platform\": \"Linux-7.0.0-1006-aws-x86_64-with-glibc2.31\",\n  \"python_implementation\": \"CPython\",\n  \"python_version\": \"3.10.18\",\n  \"server_datetime_utc\": \"2026-07-12T08:09:43.555055+00:00\",\n  \"whoami\": \"root\",\n  \"whoami_returncode\": 0,\n  \"whoami_stderr\": \"\"\n}\nPROOF_OUTPUT_END\nRESULT=AFFECTED\nEXPERIMENT_METADATA_CLEANUP_STATUS=200\nCLEANUP_NOTE=discard the temporary test instance or remove SERVER_CLEANUP_PATH on the owned host\n```\n\nThe proof file was also verified directly inside the disposable container:\n\n```bash\ncat /usr/local/lib/python3.10/site-packages/mlflow/genai/scorers/trulens/9a2d665eb38940b68983f0586f623115/artifacts/execution-proof-1783843760683972103.json\n```\n\nThe file contained:\n\n```json\n{\n  \"cwd\": \"/\",\n  \"machine\": \"x86_64\",\n  \"marker\": \"MLFLOW_REMOTE_SAFE_MARKER_1783843760683972103\",\n  \"os_release\": \"PRETTY_NAME=\\\"Debian GNU/Linux 11 (bullseye)\\\"\\nNAME=\\\"Debian GNU/Linux\\\"\\nVERSION_ID=\\\"11\\\"\\nVERSION=\\\"11 (bullseye)\\\"\\nVERSION_CODENAME=bullseye\\nID=debian\\nHOME_URL=\\\"https://www.debian.org/\\\"\\nSUPPORT_URL=\\\"https://www.debian.org/support\\\"\\nBUG_REPORT_URL=\\\"https://bugs.debian.org/\\\"\\n\",\n  \"platform\": \"Linux-7.0.0-1006-aws-x86_64-with-glibc2.31\",\n  \"python_implementation\": \"CPython\",\n  \"python_version\": \"3.10.18\",\n  \"server_datetime_utc\": \"2026-07-12T08:09:43.555055+00:00\",\n  \"whoami\": \"root\",\n  \"whoami_returncode\": 0,\n  \"whoami_stderr\": \"\"\n}\n```\n\nThe results demonstrate that:\n\n* the attacker-controlled module was uploaded beneath an allowed scorer package;\n* MLflow imported the request-selected module and invoked its constructor;\n* the constructor executed as the MLflow server process, which ran as `root` in the reproduced official image;\n* the constructor created a server-side proof file containing a random marker and bounded runtime information;\n* the proof file was retrieved successfully through MLflow\u2019s artifact-download endpoint; and\n* the later HTTP 500 occurred only after constructor execution.\n\nThe authoritative reproduction indicators are:\n\n```text\nUPLOAD_STATUS=200\nPROOF_DOWNLOAD_STATUS=200\nPROOF_MATCH=1\nRESULT=AFFECTED\n```\n\nThe disposable cloud machine and container were removed after testing.\n\nLocal Reproduction Scripts: \n[run-official-image.sh](https://github.com/user-attachments/files/29936514/run-official-image.sh)\n[poc_local.py](https://github.com/user-attachments/files/29936506/poc_local.py)",
  "id": "GHSA-26p8-2jq9-3vq9",
  "modified": "2026-10-27T21:26:32Z",
  "published": "2026-10-27T21:26:32Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/mlflow/mlflow/security/advisories/GHSA-26p8-2jq9-3vq9"
    },
    {
      "type": "WEB",
      "url": "https://github.com/mlflow/mlflow/pull/25906"
    },
    {
      "type": "WEB",
      "url": "https://github.com/mlflow/mlflow/commit/a732443e695c3c677680c54320efc394c81654ab"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/mlflow/mlflow"
    },
    {
      "type": "WEB",
      "url": "https://github.com/mlflow/mlflow/releases/tag/v3.17.0"
    }
  ],
  "schema_version": "1.4.0",
  "severity": [
    {
      "score": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H",
      "type": "CVSS_V3"
    }
  ],
  "summary": "MLflow vulnerable to remote code execution on unauthenticated servers via third-party scorer deserialization importing caller-placed artifact modules"
}