CVE-2025-55552: PyTorch integer overflow

CISO Take

PyTorch 2.8.0 crashes when combining rot90 and randn_like due to integer overflow — any inference or training pipeline using image rotation with random tensor generation is at risk of service disruption. No authentication required to trigger if the model is exposed via API. Downgrade to a stable prior release or apply vendor patch immediately; do not run 2.8.0 in production serving environments.

What is the risk?

High severity (CVSS 7.5) with low exploitability barrier: network-accessible, no credentials required, no user interaction. Impact is confined to availability (DoS), with no confidentiality or integrity breach. However, for AI/ML inference APIs in production, an availability-only crash is operationally severe — computer vision pipelines, data augmentation loops, and online training jobs using rotation ops are all exposed. PyTorch's ubiquity across ML stacks amplifies blast radius significantly.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
PyTorch	pip	—	No patch
100.9K OpenSSF 6.4 22.7K dependents Pushed 3d ago 11% patched ~216d to patch Full package profile →

Do you use PyTorch? You're affected.

How severe is it?

CVSS 3.1

7.5 / 10

EPSS

0.4%

chance of exploitation in 30 days

Higher than 31% 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 None

UI None

S Unchanged

C None

I None

A High

What should I do?

5 steps

Immediate

Downgrade to PyTorch ≤2.7.x until a patched release is available — no patch version is confirmed yet (patched field is N/A).
Workaround

Audit all code paths for concurrent use of torch.rot90 and torch.randn_like; refactor to decouple or sequence these operations.
Input validation

If rotation ops are triggered by user-controlled parameters in inference APIs, add input shape/type validation as a defense-in-depth layer.
Detection

Monitor for unexpected process crashes or OOM errors in PyTorch serving containers — set up alerting on abnormal worker restarts.
Supply chain

Pin PyTorch version in CI/CD lockfiles and block auto-upgrades to 2.8.0 via dependency management tools (pip constraints, conda lock).

What does CISA's SSVC say?

Decision Track*

Exploitation poc

Automatable Yes

Technical Impact partial

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

How is it classified?

DoS Supply Chain 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 of high-risk AI systems Article 9 - Risk management system

ISO 42001

A.6.2 - AI system risk management A.9.3 - Availability and resilience of AI systems

NIST AI RMF

GOVERN 4.2 - Organizational teams are committed to AI risk management MANAGE 2.2 - Mechanisms to sustain the value of AI systems over time

Frequently Asked Questions

What is CVE-2025-55552?

PyTorch 2.8.0 crashes when combining rot90 and randn_like due to integer overflow — any inference or training pipeline using image rotation with random tensor generation is at risk of service disruption. No authentication required to trigger if the model is exposed via API. Downgrade to a stable prior release or apply vendor patch immediately; do not run 2.8.0 in production serving environments.

Is CVE-2025-55552 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2025-55552, increasing the risk of exploitation.

How to fix CVE-2025-55552?

1. **Immediate**: Downgrade to PyTorch ≤2.7.x until a patched release is available — no patch version is confirmed yet (patched field is N/A). 2. **Workaround**: Audit all code paths for concurrent use of torch.rot90 and torch.randn_like; refactor to decouple or sequence these operations. 3. **Input validation**: If rotation ops are triggered by user-controlled parameters in inference APIs, add input shape/type validation as a defense-in-depth layer. 4. **Detection**: Monitor for unexpected process crashes or OOM errors in PyTorch serving containers — set up alerting on abnormal worker restarts. 5. **Supply chain**: Pin PyTorch version in CI/CD lockfiles and block auto-upgrades to 2.8.0 via dependency management tools (pip constraints, conda lock).

What systems are affected by CVE-2025-55552?

This vulnerability affects the following AI/ML architecture patterns: model inference serving, training pipelines, data augmentation pipelines, computer vision deployments.

What is the CVSS score for CVE-2025-55552?

CVE-2025-55552 has a CVSS v3.1 base score of 7.5 (HIGH). The EPSS exploitation probability is 0.39%.

What is the AI security impact?

Affected AI Architectures

model inference servingtraining pipelinesdata augmentation pipelinescomputer vision deployments

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, Article 9

ISO 42001: A.6.2, A.9.3

NIST AI RMF: GOVERN 4.2, MANAGE 2.2

What are the technical details?

Original Advisory

pytorch v2.8.0 was discovered to display unexpected behavior when the components torch.rot90 and torch.randn_like are used together.

Exploitation Scenario

An adversary targets a computer vision inference API built on PyTorch 2.8.0 — for example, a fraud detection service or medical imaging classifier. By submitting a crafted image request that causes the backend to invoke torch.rot90 followed by torch.randn_like (common in test-time augmentation or TTA ensembling), the adversary triggers an integer overflow crash, taking down the worker process. In a serverless or auto-scaling environment this could generate cascading restarts. In a Kubernetes cluster with a single replica, it results in service unavailability. No ML expertise is required — knowledge of the API's augmentation pipeline and the public GitHub issue is sufficient to reproduce.

Weaknesses (CWE)

CWE-190 Integer Overflow or Wraparound Primary CWE-682 Incorrect Calculation

CWE-190 — Integer Overflow or Wraparound: The product performs a calculation that can produce an integer overflow or wraparound when the logic assumes that the resulting value will always be larger than the original value. This occurs when an integer value is incremented to a value that is too large to store in the associated representation. When this occurs, the value may become a very small or negative number.

[Requirements] Ensure that all protocols are strictly defined, such that all out-of-bounds behavior can be identified simply, and require strict conformance to the protocol.
[Requirements] Use a language that does not allow this weakness to occur or provides constructs that make this weakness easier to avoid. If possible, choose a language or compiler that performs automatic bounds checking.

Source: MITRE CWE corpus.