CVE-2022-41909 — HIGH (CVSS 7.5) AI Security Vulnerability

CISO Take

Any TensorFlow serving endpoint that exposes CompositeTensor operations can be crashed remotely with no authentication by sending a malformed tensor—pure availability kill. Patch immediately to TF 2.11, 2.10.1, 2.9.3, or 2.8.4; if patching is blocked, place an API gateway in front that validates tensor types before they reach the raw ops layer. Not actively exploited in the wild, but the bar to exploit is trivially low.

Risk Assessment

High risk for organizations running TensorFlow inference APIs exposed to untrusted networks. CVSS 7.5 reflects network-accessible, zero-auth, zero-interaction exploitation path with full availability impact. The attack requires only sending a crafted invalid CompositeTensorVariant—no ML knowledge needed. Risk is amplified in multi-tenant or public-facing model serving environments where a single crash can deny service to all users. Not in CISA KEV and no evidence of active exploitation reduces immediate urgency, but the trivial exploitability warrants urgent patching during next maintenance window.

Affected Systems

Package	Ecosystem	Vulnerable Range	Patched
tensorflow	pip	—	No patch
195.0K OpenSSF 7.2 3.7K dependents Pushed today 4% patched ~1372d to patch Full package profile →

Do you use tensorflow? You're affected.

Severity & Risk

CVSS 3.1

7.5 / 10

EPSS

0.5%

chance of exploitation in 30 days

Higher than 64% 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, CISA SSVC, EPSS, trickest/cve, and Nuclei templates.

Attack Surface

AV Network

AC Low

PR None

UI None

S Unchanged

C None

I None

A High

Recommended Action

5 steps

PATCH

Upgrade TensorFlow to 2.11, 2.10.1, 2.9.3, or 2.8.4. Commits bf594d08 and 660ce5a8 contain the fix.
VALIDATE

Add input validation at the API boundary—reject requests containing malformed or unexpected tensor types before they reach raw ops. TensorFlow's tf.experimental.ExtensionType validation can help.
ISOLATE

Run inference workers as separate processes (not shared) so a crash does not cascade. Use process supervisors (systemd, Kubernetes liveness probes) to auto-restart.
RATE-LIMIT: Throttle inference API calls per client to reduce crash-loop DoS surface.
DETECT

Alert on unexpected TensorFlow process exits (signal 11 / SIGSEGV). Monitor crash dumps for CompositeTensorVariantToComponents stack frames.

CISA SSVC Assessment

Decision Track*

Exploitation poc

Automatable No

Technical Impact partial

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

Classification

DoS Framework Inference AML.T0029 - Denial of AI Service AML.T0043 - Craft Adversarial Data AML.T0049 - Exploit Public-Facing Application

Compliance Impact

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, Robustness and Cybersecurity

ISO 42001

A.6.2.5 - AI System Availability and Resilience

NIST AI RMF

MANAGE 2.2 - Mechanisms to sustain the value of deployed AI while minimizing negative impacts

OWASP LLM Top 10

LLM04 - Model Denial of Service

Frequently Asked Questions

What is CVE-2022-41909?

Any TensorFlow serving endpoint that exposes CompositeTensor operations can be crashed remotely with no authentication by sending a malformed tensor—pure availability kill. Patch immediately to TF 2.11, 2.10.1, 2.9.3, or 2.8.4; if patching is blocked, place an API gateway in front that validates tensor types before they reach the raw ops layer. Not actively exploited in the wild, but the bar to exploit is trivially low.

Is CVE-2022-41909 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2022-41909, increasing the risk of exploitation.

How to fix CVE-2022-41909?

1. PATCH: Upgrade TensorFlow to 2.11, 2.10.1, 2.9.3, or 2.8.4. Commits bf594d08 and 660ce5a8 contain the fix. 2. VALIDATE: Add input validation at the API boundary—reject requests containing malformed or unexpected tensor types before they reach raw ops. TensorFlow's tf.experimental.ExtensionType validation can help. 3. ISOLATE: Run inference workers as separate processes (not shared) so a crash does not cascade. Use process supervisors (systemd, Kubernetes liveness probes) to auto-restart. 4. RATE-LIMIT: Throttle inference API calls per client to reduce crash-loop DoS surface. 5. DETECT: Alert on unexpected TensorFlow process exits (signal 11 / SIGSEGV). Monitor crash dumps for CompositeTensorVariantToComponents stack frames.

What systems are affected by CVE-2022-41909?

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

What is the CVSS score for CVE-2022-41909?

CVE-2022-41909 has a CVSS v3.1 base score of 7.5 (HIGH). The EPSS exploitation probability is 0.46%.

Technical Details

NVD Description

TensorFlow is an open source platform for machine learning. An input `encoded` that is not a valid `CompositeTensorVariant` tensor will trigger a segfault in `tf.raw_ops.CompositeTensorVariantToComponents`. We have patched the issue in GitHub commits bf594d08d377dc6a3354d9fdb494b32d45f91971 and 660ce5a89eb6766834bdc303d2ab3902aef99d3d. The fix will be included in TensorFlow 2.11. We will also cherrypick this commit on TensorFlow 2.10.1, 2.9.3, and TensorFlow 2.8.4, as these are also affected and still in supported range.

Exploitation Scenario

An adversary identifies a publicly-accessible TensorFlow Serving endpoint (e.g., via Shodan or direct enumeration of ML inference ports). They craft an HTTP request to the prediction API containing a tensor with type=DT_VARIANT but with an invalid or missing CompositeTensorVariant encoding. When TensorFlow attempts to decompose the composite tensor via tf.raw_ops.CompositeTensorVariantToComponents, it dereferences a null pointer and segfaults—crashing the serving process. The adversary scripts this in a loop to prevent auto-recovery, achieving sustained denial of service against the ML inference layer without any credentials or special knowledge.