CVE-2022-35969: TensorFlow: DoS via malformed Conv2DBackpropInput

HIGHPoC AVAILABLE

Published September 16, 2022

CISO Take

A remotely exploitable denial-of-service exists in TensorFlow's Conv2DBackpropInput operation, triggered by passing a non-4D tensor as input_sizes—crashing the process with no authentication required. Patch to TensorFlow 2.10.0, 2.9.1, 2.8.1, or 2.7.2 immediately; no workaround exists. Exposure is limited to systems with network-accessible APIs that invoke backpropagation operations, primarily training infrastructure and fine-tuning endpoints.

What is the risk?

High risk for organizations exposing TensorFlow training or fine-tuning APIs over the network. CVSS 7.5 reflects zero-auth network exploitation with low complexity—an attacker only needs to craft a tensor with incorrect dimensions. Risk is significantly reduced for pure inference deployments since Conv2DBackpropInput is a gradient operation not invoked during forward-pass inference. Training clusters, ML experimentation platforms, and fine-tuning APIs face direct exposure and should be treated as priority remediation targets.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
TensorFlow	pip	—	No patch
201.3KOpenSSF 7.63.4K dependentsPushed 6d ago4% patched~1372d to patchFull package profile →

Do you use TensorFlow? You're affected.

How severe is it?

CVSS 3.1

7.5 / 10

EPSS

0.5%

chance of exploitation in 30 days

Higher than 40% 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?

AVNetwork

ACLow

PRNone

UINone

SUnchanged

CNone

INone

AHigh

What should I do?

5 steps

Patch immediately: upgrade to TensorFlow 2.10.0, 2.9.1, 2.8.1, or 2.7.2—no workaround exists per vendor advisory.
Network segmentation: training infrastructure APIs should never be public-facing; enforce network controls now if not already in place.
Input validation: enforce tensor dimension checks at API gateways before requests reach TensorFlow ops.
Detection: monitor for repeated TF process crashes or unexpected session terminations in training clusters as an indicator of active exploitation.
Audit exposure: inventory all network-accessible TensorFlow endpoints and confirm each requires authentication and is not reachable from untrusted networks.

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?

DoSFrameworkAPIAML.T0029- Denial of AI Service AML.T0034- Cost Harvesting AML.T0049- Exploit Public-Facing Application

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15-Accuracy, robustness and cybersecurity

ISO 42001

8.4-AI system design and development

NIST AI RMF

GOVERN-1.1-Organizational AI risk policiesMANAGE-2.2-Mechanisms for AI risk management

Frequently Asked Questions

What is CVE-2022-35969?

Is CVE-2022-35969 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2022-35969, increasing the risk of exploitation.

How to fix CVE-2022-35969?

1. Patch immediately: upgrade to TensorFlow 2.10.0, 2.9.1, 2.8.1, or 2.7.2—no workaround exists per vendor advisory. 2. Network segmentation: training infrastructure APIs should never be public-facing; enforce network controls now if not already in place. 3. Input validation: enforce tensor dimension checks at API gateways before requests reach TensorFlow ops. 4. Detection: monitor for repeated TF process crashes or unexpected session terminations in training clusters as an indicator of active exploitation. 5. Audit exposure: inventory all network-accessible TensorFlow endpoints and confirm each requires authentication and is not reachable from untrusted networks.

What systems are affected by CVE-2022-35969?

This vulnerability affects the following AI/ML architecture patterns: training pipelines, model fine-tuning services, distributed training clusters, ML experimentation platforms, model serving with training endpoints.

What is the CVSS score for CVE-2022-35969?

CVE-2022-35969 has a CVSS v3.1 base score of 7.5 (HIGH). The EPSS exploitation probability is 0.46%.

What is the AI security impact?

Affected AI Architectures

training pipelinesmodel fine-tuning servicesdistributed training clustersML experimentation platformsmodel serving with training endpoints

MITRE ATLAS Techniques

AML.T0029Denial of AI Service

AML.T0034Cost Harvesting

AML.T0049Exploit Public-Facing Application

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: 8.4

NIST AI RMF: GOVERN-1.1, MANAGE-2.2

What are the technical details?

Original Advisory

TensorFlow is an open source platform for machine learning. The implementation of `Conv2DBackpropInput` requires `input_sizes` to be 4-dimensional. Otherwise, it gives a `CHECK` failure which can be used to trigger a denial of service attack. We have patched the issue in GitHub commit 50156d547b9a1da0144d7babe665cf690305b33c. The fix will be included in TensorFlow 2.10.0. We will also cherrypick this commit on TensorFlow 2.9.1, TensorFlow 2.8.1, and TensorFlow 2.7.2, as these are also affected and still in supported range. There are no known workarounds for this issue.

Exploitation Scenario

An adversary identifies a TensorFlow-based fine-tuning API or ML training platform accessible over the network—common in internal MLOps platforms or shared GPU clusters. They craft a malicious request passing a 2D or 3D tensor as the input_sizes parameter to Conv2DBackpropInput instead of the required 4D shape. TensorFlow's internal CHECK macro triggers an assertion failure, aborting the process. In a Kubernetes ML training cluster, this can be sustained with repeated low-volume requests, creating a crash loop that stalls active training jobs, wastes expensive GPU compute time, and potentially corrupts checkpoint state if the crash occurs during a write. No exploit code or AI/ML expertise is required—only knowledge of the TensorFlow op signature.

Weaknesses (CWE)

CWE-617Reachable AssertionPrimary CWE-617Reachable Assertion

CWE-617 — Reachable Assertion: The product contains an assert() or similar statement that can be triggered by an attacker, which leads to an application exit or other behavior that is more severe than necessary.

[Implementation] If the assertion statement is required (e.g., for security or reliability), change the invoking code to avoid the conditions that trigger the assertion.
[Implementation] Consider changing the assertion to some other less "fatal" error. This might not be easy to implement, since other code may be needed to handle the error.

Source: MITRE CWE corpus.