CVE-2022-23579: TensorFlow: DoS via Grappler optimizer CHECK failure

MEDIUMPoC AVAILABLECISA: TRACK*

Published February 4, 2022

CISO Take

A low-privileged attacker can crash TensorFlow processes by submitting a crafted SavedModel that triggers an assertion failure in the Grappler optimizer. This is a real threat in multi-tenant ML platforms where users can upload or share models. Patch to TF 2.8.0, 2.7.1, 2.6.3, or 2.5.3 immediately; if patching is delayed, restrict who can submit SavedModels to your inference infrastructure.

What is the risk?

Medium risk overall, but elevated in environments exposing model-loading APIs to external or low-trust users. The low attack complexity and no user-interaction requirement make this reliably exploitable by any authenticated user. Impact is limited to availability (process crash/restart), with no data exfiltration or integrity risk. Multi-tenant ML serving platforms and shared training infrastructure carry the highest exposure.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
TensorFlow	pip	—	No patch
201.3KOpenSSF 7.63.4K dependentsPushed 5d ago4% patched~1372d to patchFull package profile →

Do you use TensorFlow? You're affected.

How severe is it?

CVSS 3.1

6.5 / 10

EPSS

0.8%

chance of exploitation in 30 days

Higher than 58% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Moderate

Exploitation Confidence

medium

○CISA SSVC: Public PoC

○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?

AVNetwork

ACLow

PRLow

UINone

SUnchanged

CNone

INone

AHigh

What should I do?

5 steps

Patch: Upgrade TensorFlow to 2.8.0, 2.7.1, 2.6.3, or 2.5.3.
Workaround: Disable Grappler optimization for untrusted SavedModels via RewriterConfig (set all optimizer levels to OFF).
Input validation: Validate SavedModel provenance and restrict upload capabilities to authenticated, trusted users only.
Isolation: Run model loading in isolated processes or containers so a crash does not impact the main serving fleet.
Detection: Alert on unexpected TF process crashes or CHECK failure logs containing 'SafeToRemoveIdentity' or 'dependency_optimizer'.

What does CISA's SSVC say?

DecisionTrack*

Exploitationpoc

AutomatableNo

Technical Impactpartial

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

How is it classified?

DoSSupply ChainFrameworkModelAML.T0010.001- AI Software AML.T0011.000- Unsafe AI Artifacts AML.T0029- Denial of AI Service

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15-Accuracy, Robustness and Cybersecurity

ISO 42001

8.4-AI System Operation and Monitoring

NIST AI RMF

MANAGE-2.2-Risks from third-party entities are managed

OWASP LLM Top 10

LLM05-Supply Chain Vulnerabilities

Frequently Asked Questions

What is CVE-2022-23579?

Is CVE-2022-23579 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2022-23579, increasing the risk of exploitation.

How to fix CVE-2022-23579?

1. Patch: Upgrade TensorFlow to 2.8.0, 2.7.1, 2.6.3, or 2.5.3. 2. Workaround: Disable Grappler optimization for untrusted SavedModels via RewriterConfig (set all optimizer levels to OFF). 3. Input validation: Validate SavedModel provenance and restrict upload capabilities to authenticated, trusted users only. 4. Isolation: Run model loading in isolated processes or containers so a crash does not impact the main serving fleet. 5. Detection: Alert on unexpected TF process crashes or CHECK failure logs containing 'SafeToRemoveIdentity' or 'dependency_optimizer'.

What systems are affected by CVE-2022-23579?

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

What is the CVSS score for CVE-2022-23579?

CVE-2022-23579 has a CVSS v3.1 base score of 6.5 (MEDIUM). The EPSS exploitation probability is 0.83%.

What is the AI security impact?

Affected AI Architectures

model servingtraining pipelinesmodel registries

MITRE ATLAS Techniques

AML.T0010.001AI Software

AML.T0011.000Unsafe AI Artifacts

AML.T0029Denial of AI Service

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: 8.4

NIST AI RMF: MANAGE-2.2

OWASP LLM Top 10: LLM05

What are the technical details?

Original Advisory

Tensorflow is an Open Source Machine Learning Framework. The Grappler optimizer in TensorFlow can be used to cause a denial of service by altering a `SavedModel` such that `SafeToRemoveIdentity` would trigger `CHECK` failures. The fix will be included in TensorFlow 2.8.0. We will also cherrypick this commit on TensorFlow 2.7.1, TensorFlow 2.6.3, and TensorFlow 2.5.3, as these are also affected and still in supported range.

Exploitation Scenario

An adversary with a low-privilege account on a shared ML platform (e.g., a data scientist account on an internal model registry or a customer on a multi-tenant ML-as-a-service) crafts a malicious TensorFlow SavedModel using publicly available exploit code from the referenced GitHub PoC. They submit this model via the platform's model upload API. When the platform's inference service or CI validation pipeline loads the model, TensorFlow's Grappler dependency optimizer processes the malicious graph, hits the CHECK failure in SafeToRemoveIdentity, and the process terminates. Repeated submissions create a sustained DoS against the model serving infrastructure.

Weaknesses (CWE)

CWE-617Reachable 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] If the assertion statement is required (e.g., for security or reliability), change the invoking code to avoid the conditions that trigger the assertion.
[Implementation] Consider changing the assertion to some other less "fatal" error. This might not be easy to implement, since other code may be needed to handle the error.

Source: MITRE CWE corpus.