CVE-2022-23586: TensorFlow SavedModel DoS crashes

CISO Take

Any TensorFlow deployment that loads external or user-supplied SavedModel files is at risk of a denial-of-service crash. Patch to TF 2.8.0, 2.7.1, 2.6.3, or 2.5.3 immediately. Enforce strict provenance controls on SavedModels—do not load models from untrusted sources without process isolation.

What is the risk?

Medium severity (CVSS 6.5) but operationally significant for ML serving infrastructure. Low attack complexity with no user interaction required makes this trivial to trigger by any authenticated user. Risk escalates substantially in multi-tenant ML platforms or model hubs where users can upload SavedModels—a single malicious file causes a full crash. No active exploitation reported; 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 3d ago 4% patched ~1372d to patch Full 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 51% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Trivial

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?

AV Network

AC Low

PR Low

UI None

S Unchanged

C None

I None

A High

What should I do?

5 steps

Patch: Upgrade to TensorFlow 2.8.0, 2.7.1, 2.6.3, or 2.5.3 per your supported branch.
Restrict model sources: Implement cryptographic provenance checks—only load SavedModels from signed, internal repositories.
Sandbox loading: Run model ingestion and validation in isolated processes or containers so a crash does not affect the production inference path.
Audit exposure: Identify all environments accepting external SavedModel uploads and prioritize patching those first.
Monitor: Alert on unexpected Python interpreter crashes in ML serving processes as an indicator of exploitation attempts.

What does CISA's SSVC say?

Decision Track*

Exploitation poc

Automatable No

Technical Impact partial

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

How is it classified?

DoS Framework Model AML.T0010.003 - Model AML.T0011.000 - Unsafe AI Artifacts AML.T0018 - Manipulate AI Model 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

Clause 8.4 - AI System Operational Controls

NIST AI RMF

MANAGE 2.4 - Risks and other undesirable impacts of AI are managed

Frequently Asked Questions

What is CVE-2022-23586?

Any TensorFlow deployment that loads external or user-supplied SavedModel files is at risk of a denial-of-service crash. Patch to TF 2.8.0, 2.7.1, 2.6.3, or 2.5.3 immediately. Enforce strict provenance controls on SavedModels—do not load models from untrusted sources without process isolation.

Is CVE-2022-23586 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2022-23586, increasing the risk of exploitation.

How to fix CVE-2022-23586?

1. Patch: Upgrade to TensorFlow 2.8.0, 2.7.1, 2.6.3, or 2.5.3 per your supported branch. 2. Restrict model sources: Implement cryptographic provenance checks—only load SavedModels from signed, internal repositories. 3. Sandbox loading: Run model ingestion and validation in isolated processes or containers so a crash does not affect the production inference path. 4. Audit exposure: Identify all environments accepting external SavedModel uploads and prioritize patching those first. 5. Monitor: Alert on unexpected Python interpreter crashes in ML serving processes as an indicator of exploitation attempts.

What systems are affected by CVE-2022-23586?

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

What is the CVSS score for CVE-2022-23586?

CVE-2022-23586 has a CVSS v3.1 base score of 6.5 (MEDIUM). The EPSS exploitation probability is 0.79%.

What is the AI security impact?

Affected AI Architectures

model servingtraining pipelinesMLOps platformsmodel registries

MITRE ATLAS Techniques

AML.T0010.003 Model

AML.T0011.000 Unsafe AI Artifacts

AML.T0018 Manipulate AI Model

AML.T0029 Denial of AI Service

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: Clause 8.4

NIST AI RMF: MANAGE 2.4

What are the technical details?

Original Advisory

Tensorflow is an Open Source Machine Learning Framework. A malicious user can cause a denial of service by altering a `SavedModel` such that assertions in `function.cc` would be falsified and crash the Python interpreter. 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 low-privilege access to an ML platform—a registered user of an internal model hub, a federated learning participant, or a compromised CI/CD pipeline—uploads a specially crafted SavedModel file. When the platform's TensorFlow runtime loads the model for validation, registry ingestion, or inference deployment, assertions in function.cc are falsified and the Python interpreter crashes. In a serving context, the inference pod goes down, causing service disruption. The attacker repeats the upload to maintain a persistent DoS condition against the endpoint without needing elevated privileges.

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.