CVE-2025-55554: PyTorch integer overflow

CISO Take

CVE-2025-55554 affects PyTorch 2.8.0 with an integer overflow in torch.nan_to_num when casting to long, exploitable remotely with no authentication to crash inference services. Impact is availability-only — no data exposure or code execution. If your ML serving infrastructure runs PyTorch 2.8.0, assess exposure of inference endpoints and prioritize patching when upstream releases a fix.

What is the risk?

Medium risk with narrow exploitability window specific to PyTorch 2.8.0. CVSS 5.3 (AV:N/AC:L/PR:N/UI:N) means it can be triggered over the network without credentials, but blast radius is limited to service availability. Not in CISA KEV. No evidence of active exploitation. Risk elevates significantly for organizations exposing PyTorch inference APIs directly to untrusted networks or user-controlled inputs that flow into numerical preprocessing stages.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
PyTorch	pip	—	No patch
100.9K OpenSSF 6.4 22.7K dependents Pushed 2d ago 11% patched ~216d to patch Full package profile →

Do you use PyTorch? You're affected.

How severe is it?

CVSS 3.1

5.3 / 10

EPSS

0.3%

chance of exploitation in 30 days

Higher than 21% 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 Network

AC Low

PR None

UI None

S Unchanged

C None

I None

A Low

What should I do?

6 steps

Inventory all deployments using PyTorch 2.8.0 and identify services exposing inference to external inputs.
Monitor https://github.com/pytorch/pytorch/issues/151510 for official patch release.
Workaround: add input validation to reject or clamp extreme numerical values before calling nan_to_num; avoid .long() conversion on untrusted float tensors without bounds checking.
For production serving, wrap inference calls in exception handlers to prevent process termination on overflow.
Consider pinning to a patched version as soon as one is released.
Detection: monitor inference service crash logs for unexpected termination correlated with nan_to_num stack traces.

What does CISA's SSVC say?

Decision Track

Exploitation none

Automatable No

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 9 - Risk management system for high-risk AI

ISO 42001

6.1.2 - AI risk assessment 8.4 - AI system documentation and operational controls

NIST AI RMF

MANAGE-2.2 - Mechanisms to sustain the value of deployed AI are evaluated and applied

Frequently Asked Questions

What is CVE-2025-55554?

CVE-2025-55554 affects PyTorch 2.8.0 with an integer overflow in torch.nan_to_num when casting to long, exploitable remotely with no authentication to crash inference services. Impact is availability-only — no data exposure or code execution. If your ML serving infrastructure runs PyTorch 2.8.0, assess exposure of inference endpoints and prioritize patching when upstream releases a fix.

Is CVE-2025-55554 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2025-55554, increasing the risk of exploitation.

How to fix CVE-2025-55554?

1. Inventory all deployments using PyTorch 2.8.0 and identify services exposing inference to external inputs. 2. Monitor https://github.com/pytorch/pytorch/issues/151510 for official patch release. 3. Workaround: add input validation to reject or clamp extreme numerical values before calling nan_to_num; avoid .long() conversion on untrusted float tensors without bounds checking. 4. For production serving, wrap inference calls in exception handlers to prevent process termination on overflow. 5. Consider pinning to a patched version as soon as one is released. 6. Detection: monitor inference service crash logs for unexpected termination correlated with nan_to_num stack traces.

What systems are affected by CVE-2025-55554?

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

What is the CVSS score for CVE-2025-55554?

CVE-2025-55554 has a CVSS v3.1 base score of 5.3 (MEDIUM). The EPSS exploitation probability is 0.29%.

What is the AI security impact?

Affected AI Architectures

model servingtraining pipelinesinference APIsdata preprocessing pipelines

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 9

ISO 42001: 6.1.2, 8.4

NIST AI RMF: MANAGE-2.2

What are the technical details?

Original Advisory

pytorch v2.8.0 was discovered to contain an integer overflow in the component torch.nan_to_num-.long().

Exploitation Scenario

An adversary targets a publicly accessible ML inference API (e.g., a feature scoring or anomaly detection service) built on PyTorch 2.8.0. They craft a payload containing extreme floating-point values (NaN, +Inf, or values near INT64_MAX) in the input tensor. When the service preprocesses the input using torch.nan_to_num(...).long(), the integer overflow triggers, crashing the inference worker. In a Kubernetes or containerized deployment, the crashed pod restarts automatically, but repeated requests create a sustained DoS — effectively making the AI service unavailable without needing any authentication or special privileges. This is particularly damaging in real-time scoring pipelines (fraud detection, content moderation) where availability is SLA-critical.

Weaknesses (CWE)

CWE-190 Integer Overflow or Wraparound

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.