# CVE-2026-118613: Ultralytics torch_load deserialization of untrusted data RCE

Source: https://junglewise.ai/threats/cve-2026-118613-ultralytics-torch-load-deserialization-of-untrusted-data-remote  
Publisher: Junglewise Threat Intelligence, https://junglewise.ai/threats  
Published: 2026-10-28. Updated: 2026-10-28  
License: CC BY 4.0 with a link back

CVE-2026-118613 · Severity: high · CVSS 7.8 · Published 2026-10-28

Ultralytics YOLOv8 machine learning framework contains a remote code execution vulnerability in its torch_load function that deserializes untrusted data without proper validation. An attacker can trick a user into opening a malicious file or visiting a malicious page to execute arbitrary code with the privileges of the current user, potentially leading to complete system compromise.

Vendors: [Ultralytics](https://junglewise.ai/threats/vendors/ultralytics).

## Quick answers

**Is CVE-2026-118613 exploited in the wild?** No. Junglewise Threat Intelligence has no report of CVE-2026-118613 being exploited in the wild.

**How severe is CVE-2026-118613?** High severity, with a CVSS score of 7.8, published 28 October 2026.

**Which products and versions are affected?** Ultralytics YOLOv8.

**Is there a fix for CVE-2026-118613?** No fixed version is recorded yet. Restrict user interaction with untrusted model files and malicious web pages, and avoid loading .pt model files from untrusted sources.</anigation>
<parameter name="timeline">[{"kind": "disclosed", "at": "2026-06-18", "note": "Vulnerability reported to vendor"}, {"kind": "advisory", "at": "2026-10-28", "note": "Coordinated public release of advisory"}] If you run it, Junglewise drafts a mitigation that fits your setup and applies it once you approve, with senior security engineers behind it.

**How is CVE-2026-118613 attacked?** It is attacked with local access, in a software library.

**Am I affected by CVE-2026-118613?** Check whether you run Ultralytics YOLOv8. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

## How to fix


Restrict user interaction with untrusted model files and malicious web pages, and avoid loading .pt model files from untrusted sources.</anigation>
<parameter name="timeline">[{"kind": "disclosed", "at": "2026-06-18", "note": "Vulnerability reported to vendor"}, {"kind": "advisory", "at": "2026-10-28", "note": "Coordinated public release of advisory"}]

## Technical details

The torch_load function in Ultralytics fails to validate user-supplied data before deserialization, allowing arbitrary code execution. The vulnerability requires local attack vector and user interaction (opening a malicious file or visiting a malicious page), but once triggered, an attacker gains code execution in the context of the current user. The flaw is a classic unsafe deserialization issue in Python's pickle-based model loading.

## Affected products

- Ultralytics YOLOv8

## References

- [https://www.zerodayinitiative.com/advisories/ZDI-26-812/](https://www.zerodayinitiative.com/advisories/ZDI-26-812/)