Intel

AIKIDO-2026-799016

ultralytics is vulnerable to Path Traversal

Pre-CVE
Found by Aikido Intel before public disclosure or CVE publication.
Published Jul 17, 2026
July 17, 2026

73

CVE Score

High

Path Traversal

Are you affected? Scan for Free

TL;DR

convert_ndjson_to_yolo() in ultralytics/data/converter.py joins NDJSON split, file, and classification class-name fields directly into output image and label paths under the generated dataset directory. A crafted NDJSON record can supply traversal segments or other unsafe path components in those fields so conversion writes outside the intended dataset tree, overwriting or creating files the converter process can reach. The fix restricts split to train/val/test, rewrites each image to a collision-free index-based filename, and maps classification classes onto dense numeric directories so untrusted display names never own filesystem paths.

How do I fix this?

Upgrade the ultralytics library to the patch version.

Who does this affect?

You are affected if you are using a version that falls within the vulnerable range and your application converts or trains on untrusted NDJSON datasets via convert_ndjson_to_yolo() or model.train(data=...) with an .ndjson source.

Versions

8.0.136 - 8.4.95

Fixed in 8.4.96

Background info

ultralytics is vulnerable to Path Traversal in versions 8.0.136 - 8.4.95.

Get secure today,
quickly and for free.

Secure your code, cloud, and runtime in one central system. Connect a repo to discover what the reasoning agents find in your codebase.

No credit card required | Scan results in 32 seconds.

Trusted by 150k+ orgs
  • Revolut
  • n8n
  • Lovable
  • Deel
  • Prove
  • Omnea