CVE-2021-41210: TensorFlow heap OOB read

CISO Take

A heap out-of-bounds read in TensorFlow's SparseCountSparseOutput shape inference allows local low-privilege attackers to read sensitive memory or crash the process. Shared ML training clusters, multi-tenant Jupyter environments, and containerized inference platforms face the highest real-world exposure. Patch all TensorFlow deployments to 2.7.0, 2.6.1, 2.5.2, or 2.4.4 immediately and audit base images in ML CI/CD pipelines.

What is the risk?

CVSS 7.1 High, but operational risk is moderate for most organizations due to the local attack vector. The vulnerability becomes significantly more dangerous in shared compute environments — GPU clusters, Jupyter hubs, or SageMaker Studio — where multiple users share the same TensorFlow runtime. Attack complexity is low once local access is obtained. No CISA KEV listing and no known active exploitation at time of disclosure, but the 2021 vintage means many unpatched deployments still exist in production.

What systems are affected?

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

Do you use TensorFlow? You're affected.

How severe is it?

CVSS 3.1

7.1 / 10

EPSS

0.1%

chance of exploitation in 30 days

Higher than 4% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Moderate

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?

AV Local

AC Low

PR Low

UI None

S Unchanged

C High

I None

A High

What should I do?

6 steps

Patch: Upgrade TensorFlow to 2.7.0, 2.6.1, 2.5.2, or 2.4.4 in all environments.
Inventory: Enumerate TF versions across training clusters, CI/CD pipelines, notebook servers, and inference containers. Use 'pip show tensorflow' or container scanning tools.
Container hygiene: Update Dockerfiles and requirements.txt/pyproject.toml constraints; rebuild and redeploy affected images.
Isolation: Until patched, enforce least-privilege access on shared ML compute — restrict who can submit arbitrary TF operations.
Detection: Alert on unexpected SIGSEGV/heap corruption crashes in TF training or serving processes.
Verify: Confirm patch application with CVE scanners (Trivy, Grype) against your ML container registry.

How is it classified?

Data Extraction Code Execution DoS Framework Training Data AML.T0010.001 - AI Software AML.T0037 - Data from Local System AML.T0049 - Exploit Public-Facing Application

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Art. 9 - Risk Management System

ISO 42001

A.6.2.6 - AI system security

NIST AI RMF

MANAGE 2.2 - Mechanisms are in place and applied to sustain the value of deployed AI systems

OWASP LLM Top 10

LLM03:2025 - Supply Chain

Frequently Asked Questions

What is CVE-2021-41210?

A heap out-of-bounds read in TensorFlow's SparseCountSparseOutput shape inference allows local low-privilege attackers to read sensitive memory or crash the process. Shared ML training clusters, multi-tenant Jupyter environments, and containerized inference platforms face the highest real-world exposure. Patch all TensorFlow deployments to 2.7.0, 2.6.1, 2.5.2, or 2.4.4 immediately and audit base images in ML CI/CD pipelines.

Is CVE-2021-41210 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2021-41210, increasing the risk of exploitation.

How to fix CVE-2021-41210?

1. Patch: Upgrade TensorFlow to 2.7.0, 2.6.1, 2.5.2, or 2.4.4 in all environments. 2. Inventory: Enumerate TF versions across training clusters, CI/CD pipelines, notebook servers, and inference containers. Use 'pip show tensorflow' or container scanning tools. 3. Container hygiene: Update Dockerfiles and requirements.txt/pyproject.toml constraints; rebuild and redeploy affected images. 4. Isolation: Until patched, enforce least-privilege access on shared ML compute — restrict who can submit arbitrary TF operations. 5. Detection: Alert on unexpected SIGSEGV/heap corruption crashes in TF training or serving processes. 6. Verify: Confirm patch application with CVE scanners (Trivy, Grype) against your ML container registry.

What systems are affected by CVE-2021-41210?

This vulnerability affects the following AI/ML architecture patterns: training pipelines, model serving, notebook environments, multi-tenant ML platforms, containerized ML workloads.

What is the CVSS score for CVE-2021-41210?

CVE-2021-41210 has a CVSS v3.1 base score of 7.1 (HIGH). The EPSS exploitation probability is 0.15%.

What is the AI security impact?

Affected AI Architectures

training pipelinesmodel servingnotebook environmentsmulti-tenant ML platformscontainerized ML workloads

MITRE ATLAS Techniques

AML.T0010.001 AI Software

AML.T0037 Data from Local System

AML.T0049 Exploit Public-Facing Application

Compliance Controls Affected

EU AI Act: Art. 9

ISO 42001: A.6.2.6

NIST AI RMF: MANAGE 2.2

OWASP LLM Top 10: LLM03:2025

What are the technical details?

Original Advisory

TensorFlow is an open source platform for machine learning. In affected versions the shape inference functions for `SparseCountSparseOutput` can trigger a read outside of bounds of heap allocated array. The fix will be included in TensorFlow 2.7.0. We will also cherrypick this commit on TensorFlow 2.6.1, TensorFlow 2.5.2, and TensorFlow 2.4.4, as these are also affected and still in supported range.

Exploitation Scenario

An adversary with low-privilege access to a shared ML training cluster — a compromised data scientist account or malicious insider — crafts a Python script that calls tf.sets.count() with a specially malformed sparse tensor. During shape inference for SparseCountSparseOutput, TensorFlow reads beyond the bounds of a heap-allocated array. In a training environment, adjacent heap memory may contain sensitive batch data, gradient tensors, or authentication tokens (e.g., cloud provider credentials loaded into the training process). Repeated triggering can also crash the training job, disrupting production model pipelines and causing financial loss in long-running distributed training runs.

Weaknesses (CWE)

CWE-125 Out-of-bounds Read Primary CWE-125 Out-of-bounds Read

CWE-125 — Out-of-bounds Read: The product reads data past the end, or before the beginning, of the intended buffer.

[Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
[Architecture and Design] Use a language that provides appropriate memory abstractions.

Source: MITRE CWE corpus.