CVE-2022-29197: TensorFlow DoS — MEDIUM

CISO Take

A missing scalar validation in TensorFlow's UnsortedSegmentJoin op allows any user with local/code execution access to crash TF processes via a crafted input. The local attack vector limits blast radius, but shared ML training platforms and multi-tenant notebook environments (Jupyter, Vertex AI, SageMaker) are the primary exposure. Patch all TF deployments to 2.9.0, 2.8.1, 2.7.2, or 2.6.4 — no workaround exists.

What is the risk?

Medium risk overall, elevated in multi-tenant ML platforms. CVSS 5.5 reflects the local-only vector, but in practice any environment where external users can submit TF graphs (shared notebooks, model serving endpoints that accept raw ops, CI/CD pipelines processing user-supplied models) becomes a DoS surface with low-skill exploitation. Not in CISA KEV and no public exploit code observed, reducing urgency for air-gapped environments.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
TensorFlow	pip	—	No patch
195.8K OpenSSF 7.1 3.7K dependents Pushed 2d ago 4% patched ~1372d to patch Full package profile →

Do you use TensorFlow? You're affected.

How severe is it?

CVSS 3.1

5.5 / 10

EPSS

0.3%

chance of exploitation in 30 days

Higher than 23% 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)

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 Local

AC Low

PR Low

UI None

S Unchanged

C None

I None

A High

What should I do?

5 steps

Patch: upgrade TensorFlow to ≥2.9.0, 2.8.1 (2.8.x branch), 2.7.2 (2.7.x branch), or 2.6.4 (2.6.x branch). No configuration workaround exists.
Restrict model execution: limit which users/services can submit arbitrary TF graphs — enforce allowlisted SavedModel signatures in serving.
Isolate training workers: run TF training jobs in sandboxed containers (gVisor, Firecracker) so a crash doesn't cascade.
Monitor for CHECK-failure crashes in TF logs (tensorflow::errors::InvalidArgument) as an anomaly signal.
Inventory: audit all TF versions across training, serving, and experimentation environments — notebook servers are frequently overlooked.

What does CISA's SSVC say?

Decision Track*

Exploitation poc

Automatable No

Technical Impact partial

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

How is it classified?

DoS Framework AML.T0010.001 - AI Software AML.T0011.000 - Unsafe AI Artifacts AML.T0029 - Denial of AI Service AML.T0049 - Exploit Public-Facing Application

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Art.15 - Accuracy, robustness and cybersecurity for high-risk AI

ISO 42001

A.6.2.5 - AI system operational performance and monitoring

NIST AI RMF

MANAGE-2.2 - Mechanisms to sustain trustworthy AI operation

OWASP LLM Top 10

LLM04 - Model Denial of Service

Frequently Asked Questions

What is CVE-2022-29197?

A missing scalar validation in TensorFlow's UnsortedSegmentJoin op allows any user with local/code execution access to crash TF processes via a crafted input. The local attack vector limits blast radius, but shared ML training platforms and multi-tenant notebook environments (Jupyter, Vertex AI, SageMaker) are the primary exposure. Patch all TF deployments to 2.9.0, 2.8.1, 2.7.2, or 2.6.4 — no workaround exists.

Is CVE-2022-29197 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2022-29197, increasing the risk of exploitation.

How to fix CVE-2022-29197?

1. Patch: upgrade TensorFlow to ≥2.9.0, 2.8.1 (2.8.x branch), 2.7.2 (2.7.x branch), or 2.6.4 (2.6.x branch). No configuration workaround exists. 2. Restrict model execution: limit which users/services can submit arbitrary TF graphs — enforce allowlisted SavedModel signatures in serving. 3. Isolate training workers: run TF training jobs in sandboxed containers (gVisor, Firecracker) so a crash doesn't cascade. 4. Monitor for CHECK-failure crashes in TF logs (`tensorflow::errors::InvalidArgument`) as an anomaly signal. 5. Inventory: audit all TF versions across training, serving, and experimentation environments — notebook servers are frequently overlooked.

What systems are affected by CVE-2022-29197?

This vulnerability affects the following AI/ML architecture patterns: training pipelines, model serving, shared notebook environments, model evaluation pipelines, ML CI/CD infrastructure.

What is the CVSS score for CVE-2022-29197?

CVE-2022-29197 has a CVSS v3.1 base score of 5.5 (MEDIUM). The EPSS exploitation probability is 0.32%.

What is the AI security impact?

Affected AI Architectures

training pipelinesmodel servingshared notebook environmentsmodel evaluation pipelinesML CI/CD infrastructure

MITRE ATLAS Techniques

AML.T0010.001 AI Software

AML.T0011.000 Unsafe AI Artifacts

AML.T0029 Denial of AI Service

AML.T0049 Exploit Public-Facing Application

Compliance Controls Affected

EU AI Act: Art.15

ISO 42001: A.6.2.5

NIST AI RMF: MANAGE-2.2

OWASP LLM Top 10: LLM04

What are the technical details?

Original Advisory

TensorFlow is an open source platform for machine learning. Prior to versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4, the implementation of `tf.raw_ops.UnsortedSegmentJoin` does not fully validate the input arguments. This results in a `CHECK`-failure which can be used to trigger a denial of service attack. The code assumes `num_segments` is a scalar but there is no validation for this before accessing its value. Versions 2.9.0, 2.8.1, 2.7.2, and 2.6.4 contain a patch for this issue.

Exploitation Scenario

An attacker with access to a shared ML notebook environment (e.g., JupyterHub, Google Colab enterprise) crafts a minimal TF script calling `tf.raw_ops.UnsortedSegmentJoin` with a multi-dimensional tensor as `num_segments` instead of a scalar. When executed, TF's CHECK macro fires, triggering an abort/crash of the TF runtime process. In a shared Jupyter environment, this terminates other users' kernels. In an automated model evaluation pipeline that imports and runs user-submitted models, an adversary submits a poisoned SavedModel embedding this op call, crashing the evaluation worker and potentially delaying or disrupting model release pipelines.

Weaknesses (CWE)

CWE-20 Improper Input Validation Primary CWE-20 Improper Input Validation

CWE-20 — Improper Input Validation: The product receives input or data, but it does not validate or incorrectly validates that the input has the properties that are required to process the data safely and correctly.

[Architecture and Design] Consider using language-theoretic security (LangSec) techniques that characterize inputs using a formal language and build "recognizers" for that language. This effectively requires parsing to be a distinct layer that effectively enforces a boundary between raw input and internal data representations, instead of allowing parser code to be scattered throughout the program, where it could be subject to errors or inconsistencies that create weaknesses. [REF-1109] [REF-1110] [REF-1111]
[Architecture and Design] Use an input validation framework such as Struts or the OWASP ESAPI Validation API. Note that using a framework does not automatically address all input validation problems; be mindful of weaknesses that could arise from misusing the framework itself (CWE-1173).

Source: MITRE CWE corpus.