CVE-2023-2356: MLflow: path traversal allows unauthenticated file read

HIGH PoC AVAILABLE NUCLEI TEMPLATE CISA: ATTEND

Published April 28, 2023

CISO Take

Any internet-accessible MLflow tracking server running below 2.3.1 is fully exposed — no credentials required, single HTTP request. Patch immediately to 2.3.1+; if patching is delayed, isolate MLflow behind a VPN or internal network boundary. Audit whether your MLflow instances store cloud credentials, API keys, or model artifacts on the same filesystem.

What is the risk?

Effective risk is HIGH to CRITICAL in practice despite the CVSS 7.5 score. The combination of no authentication, no user interaction, low complexity, and network access makes this trivially weaponizable. MLflow is often deployed with broad filesystem access on servers that also store cloud provider credentials, Hugging Face tokens, and training datasets. Organizations that expose MLflow tracking servers directly (dev shortcuts, misconfigured cloud SGs) face immediate data breach risk.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
MLflow	pip	—	No patch
26.6K OpenSSF 5.6 655 dependents Pushed 4d ago 31% patched ~51d to patch Full package profile →

Do you use MLflow? You're affected.

How severe is it?

CVSS 3.1

7.5 / 10

EPSS

4.2%

chance of exploitation in 30 days

Higher than 90% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Trivial

Exploitation Confidence

medium

○ CISA SSVC: Public PoC

○ Public PoC indexed (trickest/cve)

○ Nuclei detection template available

Composite signal derived from CISA KEV, VulnCheck KEV, CISA SSVC, EPSS, Metasploit, Exploit-DB, trickest/cve, Nuclei templates, and inthewild.io exploitation reports.

What is the attack surface?

AV Network

AC Low

PR None

UI None

S Unchanged

C High

I None

A None

What should I do?

5 steps

PATCH

Upgrade MLflow to 2.3.1 or later immediately — this is the only complete fix.
NETWORK ISOLATION

If patching is delayed, restrict MLflow access to VPN/internal networks via firewall rules or security groups. Block public internet access.
LEAST PRIVILEGE

Run MLflow process with a dedicated service account with minimal filesystem permissions — no access to ~/.aws, ~/.ssh, or application secrets.
DETECTION

Search logs for requests containing '../', '%2e%2e', or URL-encoded path traversal sequences against MLflow endpoints. Alert on any request that resolves outside the MLflow artifact root.
AUDIT

Enumerate all MLflow deployments in your environment; check if any are publicly accessible via cloud provider asset inventory tools.

What does CISA's SSVC say?

Decision Attend

Exploitation poc

Automatable Yes

Technical Impact total

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Data Extraction Data Leakage Framework AML.T0007 - Discover AI Artifacts AML.T0025 - Exfiltration via Cyber Means AML.T0035 - AI Artifact Collection AML.T0049 - Exploit Public-Facing Application AML.T0055 - Unsecured Credentials

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

A.9.3 - Protection of AI system resources

NIST AI RMF

MS-2.5 - AI Risk Monitoring — System and Component Vulnerability Management

OWASP LLM Top 10

LLM05:2025 - Improper Output Handling / Insecure Design

Frequently Asked Questions

What is CVE-2023-2356?

Is CVE-2023-2356 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2023-2356, increasing the risk of exploitation.

How to fix CVE-2023-2356?

1. PATCH: Upgrade MLflow to 2.3.1 or later immediately — this is the only complete fix. 2. NETWORK ISOLATION: If patching is delayed, restrict MLflow access to VPN/internal networks via firewall rules or security groups. Block public internet access. 3. LEAST PRIVILEGE: Run MLflow process with a dedicated service account with minimal filesystem permissions — no access to ~/.aws, ~/.ssh, or application secrets. 4. DETECTION: Search logs for requests containing '../', '%2e%2e', or URL-encoded path traversal sequences against MLflow endpoints. Alert on any request that resolves outside the MLflow artifact root. 5. AUDIT: Enumerate all MLflow deployments in your environment; check if any are publicly accessible via cloud provider asset inventory tools.

What systems are affected by CVE-2023-2356?

This vulnerability affects the following AI/ML architecture patterns: MLOps platforms, model registries, training pipelines, experiment tracking systems, model serving.

What is the CVSS score for CVE-2023-2356?

CVE-2023-2356 has a CVSS v3.1 base score of 7.5 (HIGH). The EPSS exploitation probability is 4.15%.

What is the AI security impact?

Affected AI Architectures

MLOps platformsmodel registriestraining pipelinesexperiment tracking systemsmodel serving

MITRE ATLAS Techniques

AML.T0007 Discover AI Artifacts

AML.T0025 Exfiltration via Cyber Means

AML.T0035 AI Artifact Collection

AML.T0049 Exploit Public-Facing Application

AML.T0055 Unsecured Credentials

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.9.3

NIST AI RMF: MS-2.5

OWASP LLM Top 10: LLM05:2025

What are the technical details?

Original Advisory

Relative Path Traversal in GitHub repository mlflow/mlflow prior to 2.3.1.

Exploitation Scenario

An adversary scanning for exposed MLflow tracking servers (port 5000/5001, or via cloud metadata) identifies an unpatched instance. They craft a GET request to the MLflow artifact API with a relative path traversal payload (e.g., /api/2.0/mlflow/artifacts/get?path=../../.aws/credentials&run_id=<valid_run_id>). With no authentication enforced, the server returns the AWS credentials file. The attacker uses the harvested keys to access S3 buckets containing proprietary model weights, training datasets, or customer data. In a more targeted scenario, the attacker reads the MLflow database connection string or a mounted secrets file to pivot laterally into the ML infrastructure.

Weaknesses (CWE)

CWE-23 Relative Path Traversal

CWE-23 — Relative Path Traversal: The product uses external input to construct a pathname that should be within a restricted directory, but it does not properly neutralize sequences such as ".." that can resolve to a location that is outside of that directory.

[Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
[Implementation] Inputs should be decoded and canonicalized to the application's current internal representation before being validated (CWE-180). Make sure that the application does not decode the same input twice (CWE-174). Such errors could be used to bypass allowlist validation schemes by introducing dangerous inputs after they have been checked. Use a built-in path canonicalization function (such as realpath() in C) that produces the canonical version of the pathname, which effectively removes ".." sequences and symbolic links (CWE-23, CWE-59). This includes: realpath() in C getCanonicalPath() in Java GetFullPath() in ASP.NET realpath() or abs_path() in Perl realpath() in PHP

Source: MITRE CWE corpus.