CVE-2021-29561: TensorFlow DoS — MEDIUM

CISO Take

A local attacker with low privileges can crash any TensorFlow process by passing a non-scalar tensor to tf.raw_ops.LoadAndRemapMatrix, triggering an unguarded assertion. Patch to TF 2.5.0, 2.4.2, 2.3.3, 2.2.3, or 2.1.4. Primary risk is in shared ML compute environments—Jupyter hubs, multi-tenant GPU clusters—where untrusted users can submit workloads.

What is the risk?

Medium risk in isolation, elevated in shared-compute contexts. CVSS 5.5 reflects local-only exploitability, but in practice many ML training environments expose TensorFlow ops to semi-trusted users (data scientists, contractors) who could intentionally or accidentally trigger this. No remote vector exists; the blast radius is process termination, not data exfiltration or code execution. Not actively exploited in the wild and not in CISA KEV.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
TensorFlow	pip	—	No patch
195.8K OpenSSF 7.1 3.7K dependents Pushed 5d 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 9% 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?

4 steps

Patch: Upgrade to TensorFlow 2.5.0, or backported fixes in 2.4.2 / 2.3.3 / 2.2.3 / 2.1.4.
Workaround: Validate that ckpt_path is a scalar string tensor before passing to LoadAndRemapMatrix; add tf.debugging.assert_scalar() guards in code that calls this op.
Detection: Monitor for unexpected TensorFlow process crashes (exit code from CHECK failure) in training/serving logs; alert on repeated restarts of ML workload pods.
Access control: Restrict who can submit arbitrary TF graph operations on shared infrastructure—namespace isolation in Kubernetes, user quotas on ML platforms.

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.6.2.6 - AI system availability and resilience

NIST AI RMF

MANAGE 2.2 - Mechanisms are in place to start, stop, and monitor AI system operation

OWASP LLM Top 10

LLM04 - Model Denial of Service

Frequently Asked Questions

What is CVE-2021-29561?

A local attacker with low privileges can crash any TensorFlow process by passing a non-scalar tensor to tf.raw_ops.LoadAndRemapMatrix, triggering an unguarded assertion. Patch to TF 2.5.0, 2.4.2, 2.3.3, 2.2.3, or 2.1.4. Primary risk is in shared ML compute environments—Jupyter hubs, multi-tenant GPU clusters—where untrusted users can submit workloads.

Is CVE-2021-29561 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2021-29561, increasing the risk of exploitation.

How to fix CVE-2021-29561?

1. Patch: Upgrade to TensorFlow 2.5.0, or backported fixes in 2.4.2 / 2.3.3 / 2.2.3 / 2.1.4. 2. Workaround: Validate that ckpt_path is a scalar string tensor before passing to LoadAndRemapMatrix; add tf.debugging.assert_scalar() guards in code that calls this op. 3. Detection: Monitor for unexpected TensorFlow process crashes (exit code from CHECK failure) in training/serving logs; alert on repeated restarts of ML workload pods. 4. Access control: Restrict who can submit arbitrary TF graph operations on shared infrastructure—namespace isolation in Kubernetes, user quotas on ML platforms.

What systems are affected by CVE-2021-29561?

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

What is the CVSS score for CVE-2021-29561?

CVE-2021-29561 has a CVSS v3.1 base score of 5.5 (MEDIUM). The EPSS exploitation probability is 0.19%.

What is the AI security impact?

Affected AI Architectures

training pipelinesmodel servingMLOps orchestration platforms

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.6.2.6

NIST AI RMF: MANAGE 2.2

OWASP LLM Top 10: LLM04

What are the technical details?

Original Advisory

TensorFlow is an end-to-end open source platform for machine learning. An attacker can cause a denial of service by exploiting a `CHECK`-failure coming from `tf.raw_ops.LoadAndRemapMatrix`. This is because the implementation(https://github.com/tensorflow/tensorflow/blob/d94227d43aa125ad8b54115c03cece54f6a1977b/tensorflow/core/kernels/ragged_tensor_to_tensor_op.cc#L219-L222) assumes that the `ckpt_path` is always a valid scalar. However, an attacker can send any other tensor as the first argument of `LoadAndRemapMatrix`. This would cause the rank `CHECK` in `scalar<T>()()` to trigger and terminate the process. 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 adversary with low-privilege access to a shared ML compute platform (e.g., a data analyst account on a Jupyter Hub or a compromised CI/CD pipeline) submits a training job containing a call to tf.raw_ops.LoadAndRemapMatrix with a non-scalar tensor as the first argument. The CHECK assertion fires immediately, terminating the TensorFlow process. In a distributed training scenario, this kills the chief worker, forcing the entire job to restart—potentially wiping hours of compute and triggering costly re-runs. In a model-serving context where checkpoint hot-reloading is enabled, repeated triggering can keep the inference service in a restart loop, effectively achieving sustained denial of service.

Weaknesses (CWE)

CWE-617 Reachable Assertion

CWE-617 — Reachable Assertion: The product contains an assert() or similar statement that can be triggered by an attacker, which leads to an application exit or other behavior that is more severe than necessary.

[Implementation] Make sensitive open/close operation non reachable by directly user-controlled data (e.g. open/close resources)
[Implementation] Perform input validation on user data.

Source: MITRE CWE corpus.