CVE-2022-21734: TensorFlow DoS — MEDIUM

CISO Take

A low-privilege network attacker can crash any TensorFlow serving instance using the MapStage op by sending a non-scalar key tensor, triggering a process-aborting CHECK failure. Patch to TF 2.8.0, 2.7.1, 2.6.3, or 2.5.3 immediately. If patching is delayed, restrict inference API access to authenticated internal users only.

What is the risk?

Medium severity but operationally significant for production AI serving. CVSS 6.5 (AV:N/AC:L/PR:L) means any network-reachable TF inference API can be crashed with minimal skill. Only availability is impacted — no data exfiltration risk. Effective risk elevates if TF Serving is internet-exposed, lacks restart resilience, or if AI service uptime is tied to SLA obligations or regulated system operation.

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

Patch: Upgrade to TensorFlow 2.8.0 or backports 2.7.1, 2.6.3, 2.5.3.
Validate inputs: Enforce scalar key constraints at API boundaries before tensors reach MapStage.
Harden access: Require authentication (API keys, mutual TLS) on all TF Serving endpoints — PR:L means any credentialed user can trigger this.
Detect: Alert on abnormal TF serving worker restarts — repeated crashes in a short window signal active exploitation.
Containerize with limits: Set restart limits and exponential backoff in Kubernetes to prevent crash loops from amplifying DoS impact.

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.T0010.001 - AI Software 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

Article 15 - Accuracy, robustness and cybersecurity

ISO 42001

A.9.2 - AI system testing and validation

NIST AI RMF

MANAGE-2.2 - Mechanisms to sustain value of deployed AI systems

OWASP LLM Top 10

LLM04 - Model Denial of Service

Frequently Asked Questions

What is CVE-2022-21734?

A low-privilege network attacker can crash any TensorFlow serving instance using the MapStage op by sending a non-scalar key tensor, triggering a process-aborting CHECK failure. Patch to TF 2.8.0, 2.7.1, 2.6.3, or 2.5.3 immediately. If patching is delayed, restrict inference API access to authenticated internal users only.

Is CVE-2022-21734 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2022-21734, increasing the risk of exploitation.

How to fix CVE-2022-21734?

1. Patch: Upgrade to TensorFlow 2.8.0 or backports 2.7.1, 2.6.3, 2.5.3. 2. Validate inputs: Enforce scalar key constraints at API boundaries before tensors reach MapStage. 3. Harden access: Require authentication (API keys, mutual TLS) on all TF Serving endpoints — PR:L means any credentialed user can trigger this. 4. Detect: Alert on abnormal TF serving worker restarts — repeated crashes in a short window signal active exploitation. 5. Containerize with limits: Set restart limits and exponential backoff in Kubernetes to prevent crash loops from amplifying DoS impact.

What systems are affected by CVE-2022-21734?

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

What is the CVSS score for CVE-2022-21734?

CVE-2022-21734 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 servingtraining pipelinesinference APIs

MITRE ATLAS Techniques

AML.T0010.001 AI Software

AML.T0029 Denial of AI Service

AML.T0049 Exploit Public-Facing Application

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.9.2

NIST AI RMF: MANAGE-2.2

OWASP LLM Top 10: LLM04

What are the technical details?

Original Advisory

Tensorflow is an Open Source Machine Learning Framework. The implementation of `MapStage` is vulnerable a `CHECK`-fail if the key tensor is not a scalar. 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 a valid low-privilege API key to a TF Serving endpoint sends a crafted inference request with a non-scalar tensor (e.g., a 2D matrix) as the MapStage key. The MapStage kernel hits its CHECK assertion, immediately aborting the TensorFlow process. In a Kubernetes deployment without crash loop backoff, the container restarts indefinitely, creating a sustained denial-of-service that kills model availability. No ML expertise is required — only knowledge of the TFX/TF Serving proto schema, which is publicly documented. Automation is trivial: a single script can maintain continuous service disruption.

Weaknesses (CWE)

CWE-843 Access of Resource Using Incompatible Type ('Type Confusion') Primary CWE-843 Access of Resource Using Incompatible Type ('Type Confusion')

CWE-843 — Access of Resource Using Incompatible Type ('Type Confusion'): The product allocates or initializes a resource such as a pointer, object, or variable using one type, but it later accesses that resource using a type that is incompatible with the original type.

Source: MITRE CWE corpus.