CVE-2021-29578: TensorFlow heap buffer overflow

CISO Take

TensorFlow 2.1.x–2.4.x contains a heap buffer overflow in FractionalAvgPoolGrad due to missing bounds validation on pooling sequence inputs. In shared ML training environments—Jupyter hubs, GPU clusters, Kubeflow—'local access' effectively means any authenticated user, making this a credible privilege escalation path. Patch to TF 2.5.0 or the available backports immediately; isolate training workloads in separate containers as a compensating control.

What is the risk?

CVSS 7.8 High with local attack vector limits internet-facing exposure, but multi-tenant ML platforms routinely grant 'local' access to many users. No active exploitation or CISA KEV entry, and the CVE is from 2021, meaning unpatched systems have had three-plus years of exposure window. The C:H/I:H/A:H impact triad makes this a full-compromise vector within the process scope. Risk is elevated for any organization running shared training infrastructure where users can submit arbitrary model code.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
TensorFlow	pip	—	No patch
195.8K OpenSSF 7.1 3.7K dependents Pushed 3d ago 4% patched ~1372d to patch Full package profile →

Do you use TensorFlow? You're affected.

How severe is it?

CVSS 3.1

7.8 / 10

EPSS

0.2%

chance of exploitation in 30 days

Higher than 11% 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 High

A High

What should I do?

1 step

1) Upgrade to TensorFlow 2.5.0; if constrained to a prior branch, apply backports: 2.4.2, 2.3.3, 2.2.3, or 2.1.4. 2) Run training jobs in isolated single-use containers—never share a TF process across trust boundaries. 3) Restrict who can submit raw tf.raw_ops calls in shared platforms; enforce allowlists of approved ops if feasible. 4) Audit model code for direct use of tf.raw_ops.FractionalAvgPoolGrad with externally-controlled tensor shapes. 5) Ensure ASLR and stack canaries are enabled in training host OS. 6) Rotate any secrets (cloud credentials, API tokens) that were accessible in training worker environments on unpatched systems.

How is it classified?

Code Execution Framework Training Data AML.T0010.001 - AI Software 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.10.1 - AI system security and resilience

NIST AI RMF

MANAGE 2.2 - Mechanisms are in place to sustain AI risk management practices

OWASP LLM Top 10

LLM03:2025 - Supply Chain

Frequently Asked Questions

What is CVE-2021-29578?

TensorFlow 2.1.x–2.4.x contains a heap buffer overflow in FractionalAvgPoolGrad due to missing bounds validation on pooling sequence inputs. In shared ML training environments—Jupyter hubs, GPU clusters, Kubeflow—'local access' effectively means any authenticated user, making this a credible privilege escalation path. Patch to TF 2.5.0 or the available backports immediately; isolate training workloads in separate containers as a compensating control.

Is CVE-2021-29578 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2021-29578, increasing the risk of exploitation.

How to fix CVE-2021-29578?

1) Upgrade to TensorFlow 2.5.0; if constrained to a prior branch, apply backports: 2.4.2, 2.3.3, 2.2.3, or 2.1.4. 2) Run training jobs in isolated single-use containers—never share a TF process across trust boundaries. 3) Restrict who can submit raw tf.raw_ops calls in shared platforms; enforce allowlists of approved ops if feasible. 4) Audit model code for direct use of tf.raw_ops.FractionalAvgPoolGrad with externally-controlled tensor shapes. 5) Ensure ASLR and stack canaries are enabled in training host OS. 6) Rotate any secrets (cloud credentials, API tokens) that were accessible in training worker environments on unpatched systems.

What systems are affected by CVE-2021-29578?

This vulnerability affects the following AI/ML architecture patterns: training pipelines, shared ML platforms, distributed training clusters, model experimentation environments, containerized ML workloads.

What is the CVSS score for CVE-2021-29578?

CVE-2021-29578 has a CVSS v3.1 base score of 7.8 (HIGH). The EPSS exploitation probability is 0.21%.

What is the AI security impact?

Affected AI Architectures

training pipelinesshared ML platformsdistributed training clustersmodel experimentation environmentscontainerized ML workloads

MITRE ATLAS Techniques

AML.T0010.001 AI Software

AML.T0049 Exploit Public-Facing Application

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.10.1

NIST AI RMF: MANAGE 2.2

OWASP LLM Top 10: LLM03:2025

What are the technical details?

Original Advisory

TensorFlow is an end-to-end open source platform for machine learning. The implementation of `tf.raw_ops.FractionalAvgPoolGrad` is vulnerable to a heap buffer overflow. The implementation(https://github.com/tensorflow/tensorflow/blob/dcba796a28364d6d7f003f6fe733d82726dda713/tensorflow/core/kernels/fractional_avg_pool_op.cc#L216) fails to validate that the pooling sequence arguments have enough elements as required by the `out_backprop` tensor shape. The fix will be included in TensorFlow 2.5.0. We will also cherrypick this commit on TensorFlow 2.4.2, TensorFlow 2.3.3, TensorFlow 2.2.3 and TensorFlow 2.1.4, as these are also affected and still in supported range.

Exploitation Scenario

An attacker with access to a shared ML training cluster submits a crafted training job containing a model that calls tf.raw_ops.FractionalAvgPoolGrad with pooling_sequence arguments deliberately undersized relative to the out_backprop tensor shape. The missing bounds check in fractional_avg_pool_op.cc allows a write beyond the allocated heap buffer. On a vulnerable unpatched host, this can be weaponized for code execution within the training worker process—enabling the attacker to exfiltrate other users' model checkpoints, read training datasets, or harvest cloud credentials stored in environment variables, then pivot laterally within the ML infrastructure.

Weaknesses (CWE)

CWE-787 Out-of-bounds Write Primary CWE-119 Improper Restriction of Operations within the Bounds of a Memory Buffer

CWE-787 — Out-of-bounds Write: The product writes data past the end, or before the beginning, of the intended buffer.

[Requirements] Use a language that does not allow this weakness to occur or provides constructs that make this weakness easier to avoid. For example, many languages that perform their own memory management, such as Java and Perl, are not subject to buffer overflows. Other languages, such as Ada and C#, typically provide overflow protection, but the protection can be disabled by the programmer. Be wary that a language's interface to native code may still be subject to overflows, even if the language itself is theoretically safe.
[Architecture and Design] Use a vetted library or framework that does not allow this weakness to occur or provides constructs that make this weakness easier to avoid. Examples include the Safe C String Library (SafeStr) by Messier and Viega [REF-57], and the Strsafe.h library from Microsoft [REF-56]. These libraries provide safer versions of overflow-prone string-handling functions.

Source: MITRE CWE corpus.