CVE-2022-23582: TensorFlow SavedModel CHECK-fail

CISO Take

A low-privileged remote attacker can crash TensorFlow model serving by submitting a crafted SavedModel with malformed tensor shapes, triggering a fatal assertion failure. Patch to TF 2.8.0, 2.7.1, 2.6.3, or 2.5.3 immediately. If your pipelines load SavedModels from external or shared sources, validate model integrity before loading.

What is the risk?

Medium severity overall, but operationally high for teams running TensorFlow serving endpoints accessible to external or low-trust users. Exploitability is straightforward—crafting a malformed SavedModel requires only basic TensorFlow knowledge. Impact is limited to availability (no code execution, no data exfiltration), but sustained DoS against ML inference infrastructure disrupts production AI services with no data loss risk to offset downtime.

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

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

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 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 (fix commit c2426bba).
Validate SavedModels in a sandboxed process before loading into serving infrastructure.
Restrict SavedModel upload/submission to trusted principals only—do not allow anonymous or low-trust users to push models.
Monitor TF serving processes for unexpected crashes or CHECK failures in logs.
Implement process supervision (systemd, supervisord) to auto-restart crashed serving instances and reduce MTTR.

How is it classified?

DoS Framework Inference AML.T0011.000 - Unsafe AI Artifacts 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.6.1 - AI risk management

NIST AI RMF

MANAGE 2.2 - Mechanisms for managing AI risks

OWASP LLM Top 10

LLM10:2025 - Unbounded Consumption

Frequently Asked Questions

What is CVE-2022-23582?

A low-privileged remote attacker can crash TensorFlow model serving by submitting a crafted SavedModel with malformed tensor shapes, triggering a fatal assertion failure. Patch to TF 2.8.0, 2.7.1, 2.6.3, or 2.5.3 immediately. If your pipelines load SavedModels from external or shared sources, validate model integrity before loading.

Is CVE-2022-23582 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2022-23582, increasing the risk of exploitation.

How to fix CVE-2022-23582?

1. Patch: Upgrade to TensorFlow 2.8.0, 2.7.1, 2.6.3, or 2.5.3 (fix commit c2426bba). 2. Validate SavedModels in a sandboxed process before loading into serving infrastructure. 3. Restrict SavedModel upload/submission to trusted principals only—do not allow anonymous or low-trust users to push models. 4. Monitor TF serving processes for unexpected crashes or CHECK failures in logs. 5. Implement process supervision (systemd, supervisord) to auto-restart crashed serving instances and reduce MTTR.

What systems are affected by CVE-2022-23582?

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

What is the CVSS score for CVE-2022-23582?

CVE-2022-23582 has a CVSS v3.1 base score of 6.5 (MEDIUM). The EPSS exploitation probability is 0.77%.

What is the AI security impact?

Affected AI Architectures

model servingtraining pipelinesML inference infrastructure

MITRE ATLAS Techniques

AML.T0011.000 Unsafe AI Artifacts

AML.T0029 Denial of AI Service

AML.T0049 Exploit Public-Facing Application

Compliance Controls Affected

EU AI Act: Article 15

ISO 42001: A.6.1

NIST AI RMF: MANAGE 2.2

OWASP LLM Top 10: LLM10:2025

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 `TensorByteSize` would trigger `CHECK` failures. `TensorShape` constructor throws a `CHECK`-fail if shape is partial or has a number of elements that would overflow the size of an `int`. The `PartialTensorShape` constructor instead does not cause a `CHECK`-abort if the shape is partial, which is exactly what this function needs to be able to return `-1`. 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 attacker with low-privilege authenticated access—a registered API user, internal employee, or compromised account—submits a crafted SavedModel where tensor shapes are defined as partial or sized to overflow a signed int. When the TF backend calls TensorByteSize() on this model during loading, the TensorShape constructor triggers a CHECK assertion failure, aborting the process. Without automatic restart, the ML inference endpoint goes down. The attacker can loop submissions to maintain persistent DoS against the AI serving layer.

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.