CVE-2022-21735: TensorFlow DoS — MEDIUM

CISO Take

A low-privileged remote attacker can crash TensorFlow inference processes by sending crafted inputs that trigger a division by zero in FractionalMaxPool. If your ML serving infrastructure exposes TensorFlow endpoints over the network, this is a practical availability threat with no exploit complexity barrier. Patch to TensorFlow 2.8.0, 2.7.1, 2.6.3, or 2.5.3 immediately.

What is the risk?

Medium severity but operationally significant for ML serving environments. Network-exploitable with low complexity and only low-privilege access means the attack surface is broad. No confidentiality or integrity impact, but process termination of an inference server causes full service interruption — especially disruptive in unattended or high-availability serving setups without automated recovery.

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

6.5 / 10

EPSS

0.8%

chance of exploitation in 30 days

Higher than 51% 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 Network

AC Low

PR Low

UI None

S Unchanged

C None

I None

A High

What should I do?

5 steps

Upgrade TensorFlow to 2.8.0, 2.7.1, 2.6.3, or 2.5.3 (official backports available).
If immediate upgrade is blocked, add input validation to reject tensor dimensions that produce zero-value pooling parameters.
Deploy process supervisors (systemd, Docker restart policies, Kubernetes liveness probes) to auto-restart crashed inference processes.
Restrict network access to TF serving APIs to authenticated, trusted callers only — do not expose raw inference APIs to the public internet.
Monitor for unexpected process terminations in inference infrastructure as a detection signal.

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 Inference 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

ISO 42001

8.4 - AI system operation and monitoring

NIST AI RMF

RV 2.2 - Evaluate AI system robustness and resilience

Frequently Asked Questions

What is CVE-2022-21735?

A low-privileged remote attacker can crash TensorFlow inference processes by sending crafted inputs that trigger a division by zero in FractionalMaxPool. If your ML serving infrastructure exposes TensorFlow endpoints over the network, this is a practical availability threat with no exploit complexity barrier. Patch to TensorFlow 2.8.0, 2.7.1, 2.6.3, or 2.5.3 immediately.

Is CVE-2022-21735 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2022-21735, increasing the risk of exploitation.

How to fix CVE-2022-21735?

1. Upgrade TensorFlow to 2.8.0, 2.7.1, 2.6.3, or 2.5.3 (official backports available). 2. If immediate upgrade is blocked, add input validation to reject tensor dimensions that produce zero-value pooling parameters. 3. Deploy process supervisors (systemd, Docker restart policies, Kubernetes liveness probes) to auto-restart crashed inference processes. 4. Restrict network access to TF serving APIs to authenticated, trusted callers only — do not expose raw inference APIs to the public internet. 5. Monitor for unexpected process terminations in inference infrastructure as a detection signal.

What systems are affected by CVE-2022-21735?

This vulnerability affects the following AI/ML architecture patterns: model serving, inference, training pipelines.

What is the CVSS score for CVE-2022-21735?

CVE-2022-21735 has a CVSS v3.1 base score of 6.5 (MEDIUM). The EPSS exploitation probability is 0.77%.

What is the AI security impact?

Affected AI Architectures

model servinginferencetraining pipelines

MITRE ATLAS Techniques

AML.T0029 Denial of AI Service

AML.T0049 Exploit Public-Facing Application

Compliance Controls Affected

EU AI Act: Art. 15

ISO 42001: 8.4

NIST AI RMF: RV 2.2

What are the technical details?

Original Advisory

Tensorflow is an Open Source Machine Learning Framework. The implementation of `FractionalMaxPool` can be made to crash a TensorFlow process via a division by 0. The fix will be included in TensorFlow 2.8.0. We will also cherrypick this commit on TensorFlow 2.7.1, TensorFlow 2.6.3, and TensorFlow 2.5.3, as these are also affected and still in supported range.

Exploitation Scenario

An attacker with API access to a TensorFlow-based model serving endpoint submits a crafted inference request with tensor dimensions that trigger FractionalMaxPool with parameters causing a division by zero. No ML expertise is required — the attacker discovers the endpoint accepts FractionalMaxPool-backed model inputs and crafts a malformed tensor. The TF process crashes, taking down inference for all consumers sharing that server. In high-throughput serving setups without auto-recovery, this yields sustained outage with minimal attacker effort and repeatability.