CVE-2021-29618: TensorFlow DoS crash

CISO Take

This is a local denial-of-service vulnerability in TensorFlow that crashes the process when tf.transpose receives a complex tensor alongside conjugate=True. No data exfiltration or code execution is possible, but in shared ML environments (JupyterHub, MLflow, multi-tenant training clusters), a low-privileged user can intentionally crash workloads. Patch to TF 2.5.0, 2.4.2, 2.3.3, 2.2.3, or 2.1.4 immediately if running affected versions in shared or production environments.

What is the risk?

Medium risk overall, but context-dependent. Exploitation is trivial—any authenticated local user can crash a TensorFlow process with a single API call. The blast radius is limited to availability: no confidentiality or integrity impact. Risk escalates significantly in multi-tenant ML platforms where process crashes can disrupt co-running training jobs, corrupt checkpoint state, or degrade model serving SLAs. Not exploitable remotely without an adjacent code execution primitive.

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

5.5 / 10

EPSS

0.2%

chance of exploitation in 30 days

Higher than 14% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Trivial

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 None

I None

A High

What should I do?

5 steps

Upgrade TensorFlow to 2.5.0 or apply the backport patches: 2.4.2, 2.3.3, 2.2.3, 2.1.4.
If immediate patching is not possible, add an input validation layer to reject or flag requests combining complex dtype tensors with conjugate=True before they reach tf.transpose.
In multi-tenant environments, isolate TensorFlow workers per user to contain blast radius.
Monitor for unexpected TensorFlow process restarts or OOM/crash signals in serving infrastructure logs as a detection signal.
Pin TF versions in Docker images and CI pipelines to patched builds.

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.9.7 - AI system availability and resilience

NIST AI RMF

MANAGE-2.4 - Residual risks from AI systems are managed

OWASP LLM Top 10

LLM10 - Model Denial of Service

Frequently Asked Questions

What is CVE-2021-29618?

This is a local denial-of-service vulnerability in TensorFlow that crashes the process when tf.transpose receives a complex tensor alongside conjugate=True. No data exfiltration or code execution is possible, but in shared ML environments (JupyterHub, MLflow, multi-tenant training clusters), a low-privileged user can intentionally crash workloads. Patch to TF 2.5.0, 2.4.2, 2.3.3, 2.2.3, or 2.1.4 immediately if running affected versions in shared or production environments.

Is CVE-2021-29618 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2021-29618, increasing the risk of exploitation.

How to fix CVE-2021-29618?

1. Upgrade TensorFlow to 2.5.0 or apply the backport patches: 2.4.2, 2.3.3, 2.2.3, 2.1.4. 2. If immediate patching is not possible, add an input validation layer to reject or flag requests combining complex dtype tensors with conjugate=True before they reach tf.transpose. 3. In multi-tenant environments, isolate TensorFlow workers per user to contain blast radius. 4. Monitor for unexpected TensorFlow process restarts or OOM/crash signals in serving infrastructure logs as a detection signal. 5. Pin TF versions in Docker images and CI pipelines to patched builds.

What systems are affected by CVE-2021-29618?

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

What is the CVSS score for CVE-2021-29618?

CVE-2021-29618 has a CVSS v3.1 base score of 5.5 (MEDIUM). The EPSS exploitation probability is 0.23%.

What is the AI security impact?

Affected AI Architectures

training pipelinesmodel servingjupyter notebooksbatch inference

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.9.7

NIST AI RMF: MANAGE-2.4

OWASP LLM Top 10: LLM10

What are the technical details?

Original Advisory

TensorFlow is an end-to-end open source platform for machine learning. Passing a complex argument to `tf.transpose` at the same time as passing `conjugate=True` argument results in a crash. 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 local access to a shared ML platform (e.g., a data scientist account on a JupyterHub cluster or a rogue notebook job on SageMaker) submits a minimal script: `import tensorflow as tf; import numpy as np; x = tf.constant(np.array([1+2j]), dtype=tf.complex64); tf.transpose(x, conjugate=True)`. This immediately crashes the TensorFlow kernel or worker process. On a shared inference server, an adversary with API access to a model endpoint that internally calls tf.transpose on complex inputs could trigger repeated crashes to degrade availability—effectively a low-cost DoS against ML infrastructure without any exploit sophistication.