CVE-2024-21799: Intel Extension for Transformers: path traversal privesc

HIGHPoC AVAILABLE

Published November 13, 2024

CISO Take

Intel Extension for Transformers before v1.5 allows any authenticated local user to escalate privileges via path traversal—a trivial exploit on shared ML infrastructure like GPU clusters. Upgrade to v1.5+ immediately and audit who has shell access to systems running this library. The real risk is the blast radius: a compromised data scientist account becomes a root foothold on your ML training infrastructure.

What is the risk?

Medium-high risk for organizations operating shared ML compute environments. CVSS 7.1 with local attack vector, low complexity, and low privilege requirement means any user account on an affected system is a potential escalation vector. Not in CISA KEV and no confirmed public exploits, but the technique is textbook—no AI expertise required. Highest exposure in multi-tenant GPU clusters, MLOps platforms, and CI/CD pipelines where multiple users share the same compute nodes. Impact scores (I:H, A:H) reflect potential for full system compromise or destruction of ML artifacts.

How severe is it?

CVSS 3.1

7.1 / 10

EPSS

0.7%

chance of exploitation in 30 days

Higher than 55% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Trivial

Exploitation Confidence

medium

○Public PoC indexed (trickest/cve)

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?

AVLocal

ACLow

PRLow

UINone

SUnchanged

CNone

IHigh

AHigh

What should I do?

1 step

1) Patch: Upgrade Intel Extension for Transformers to v1.5+. Verify with: pip show intel-extension-for-transformers. 2) Inventory: Scan all ML servers, training nodes, and inference hosts—run: pip list --format=columns | grep intel-extension across your fleet. 3) Access control: Until patched, restrict local shell access on ML infrastructure to minimum required users; enforce SSH key-based auth only. 4) File integrity monitoring: Deploy FIM on critical directories (/etc, /root, model artifact paths) and alert on writes from Python/ML processes. 5) Container enforcement: For containerized ML workloads, enforce read-only mounts on system directories and drop DAC_OVERRIDE capabilities. 6) Secrets audit: Post-patch, rotate any credentials that may have been accessible on affected systems.

What does CISA's SSVC say?

DecisionTrack

Exploitationnone

AutomatableNo

Technical Impactpartial

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

How is it classified?

Code ExecutionSupply ChainFrameworkAML.T0010.001- AI Software AML.T0025- Exfiltration via Cyber Means AML.T0035- AI Artifact Collection AML.T0037- Data from Local System

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Art. 15-Accuracy, robustness and cybersecurity

ISO 42001

Clause 6.1-Actions to address risks and opportunities

NIST AI RMF

MANAGE 3.1-Risks and mitigations are identified and prioritized

OWASP LLM Top 10

LLM05-Supply Chain Vulnerabilities

Frequently Asked Questions

What is CVE-2024-21799?

Is CVE-2024-21799 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2024-21799, increasing the risk of exploitation.

How to fix CVE-2024-21799?

1) Patch: Upgrade Intel Extension for Transformers to v1.5+. Verify with: pip show intel-extension-for-transformers. 2) Inventory: Scan all ML servers, training nodes, and inference hosts—run: pip list --format=columns | grep intel-extension across your fleet. 3) Access control: Until patched, restrict local shell access on ML infrastructure to minimum required users; enforce SSH key-based auth only. 4) File integrity monitoring: Deploy FIM on critical directories (/etc, /root, model artifact paths) and alert on writes from Python/ML processes. 5) Container enforcement: For containerized ML workloads, enforce read-only mounts on system directories and drop DAC_OVERRIDE capabilities. 6) Secrets audit: Post-patch, rotate any credentials that may have been accessible on affected systems.

What systems are affected by CVE-2024-21799?

This vulnerability affects the following AI/ML architecture patterns: training pipelines, model serving, ML development environments, shared GPU clusters, MLOps platforms.

What is the CVSS score for CVE-2024-21799?

CVE-2024-21799 has a CVSS v3.1 base score of 7.1 (HIGH). The EPSS exploitation probability is 0.75%.

What is the AI security impact?

Affected AI Architectures

training pipelinesmodel servingML development environmentsshared GPU clustersMLOps platforms

MITRE ATLAS Techniques

AML.T0010.001AI Software

AML.T0025Exfiltration via Cyber Means

AML.T0035AI Artifact Collection

AML.T0037Data from Local System

Compliance Controls Affected

EU AI Act: Art. 15

ISO 42001: Clause 6.1

NIST AI RMF: MANAGE 3.1

OWASP LLM Top 10: LLM05

What are the technical details?

Original Advisory

Path traversal for some Intel(R) Extension for Transformers software before version 1.5 may allow an authenticated user to potentially enable escalation of privilege via local access.

Exploitation Scenario

An attacker with a low-privileged account on a shared GPU training cluster—say, a compromised data scientist credential—calls a vulnerable file operation in Intel Extension for Transformers with a crafted path such as '../../root/.ssh/authorized_keys'. The library writes attacker-controlled content to the root SSH authorized_keys file, granting the attacker persistent root SSH access. From there, the attacker has unrestricted access to all model weights, training datasets, API keys stored in .env files, and the ability to poison models or exfiltrate proprietary IP. The entire operation requires no GPU, no ML knowledge, and no exploit tooling beyond a basic path traversal string.

Weaknesses (CWE)

CWE-22Improper Limitation of a Pathname to a Restricted Directory

CWE-22 — Improper Limitation of a Pathname to a Restricted Directory: The product uses external input to construct a pathname that is intended to identify a file or directory that is located underneath a restricted parent directory, but the product does not properly neutralize special elements within the pathname that can cause the pathname to resolve to a location that is outside of the restricted 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
[Architecture and Design] For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.

Source: MITRE CWE corpus.