CVE-2026-115924: Rembg path traversal and file disclosure via sam_model parameter

CVE-2026-115924 · Severity: high · CVSS 8.6 · Published 2026-10-26

Rembg is a background removal tool that exposes a public API endpoint. An attacker can manipulate the sam_model parameter to place malicious files on the server and read arbitrary files accessible to the process. The attack exploits differences between how HTTP URLs and local file paths handle URL fragments, allowing an unauthenticated attacker to disclose sensitive files or execute arbitrary code through specially crafted ONNX neural network models.

Technologies: Danielgatis Rembg, rembg (PyPI). Vendors: Danielgatis, PyPI.

Quick answers

Is CVE-2026-115924 exploited in the wild?
No. Junglewise Threat Intelligence has no report of CVE-2026-115924 being exploited in the wild.
How severe is CVE-2026-115924?
High severity, with a CVSS score of 8.6, published 26 October 2026.
Which products and versions are affected?
Danielgatis Rembg prior to 2.0.82; rembg (PyPI) < 2.0.82, ECOSYSTEM:introduced=0,fixed=2.0.82.
Is there a fix for CVE-2026-115924?
Yes. rembg (PyPI): upgrade to 2.0.82 or later. Until then: Restrict the sam_model parameter to only the three published checkpoints before it reaches path or URL construction, and pin MD5 digests for SAM model assets to detect tampering. Junglewise finds every place you run it, drafts the upgrade and applies it once you approve.
How is CVE-2026-115924 attacked?
It is attacked over the network, in a software library.
Am I affected by CVE-2026-115924?
Check whether you run Danielgatis Rembg or rembg (PyPI) at an affected version. 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

  • rembg (PyPI): upgrade to 2.0.82 or later
    pip install "rembg>=2.0.82"

Restrict the sam_model parameter to only the three published checkpoints before it reaches path or URL construction, and pin MD5 digests for SAM model assets to detect tampering.

Technical details

The /api/remove endpoint accepts an attacker-controlled sam_model parameter from the extras payload and interpolates it directly into both download URLs and local file paths without validation. By combining path traversal syntax with URL fragments (e.g., sam_vit_b_01ec64#/../../../evil), an attacker exploits the fact that pooch.retrieve() discards URL fragments during HTTP requests while POSIX paths retain them. This allows writing malicious ONNX files outside the intended model directory; these graphs can load external data from nearby files and encode their bytes into returned masks, disclosing arbitrary readable files.

Affected products

  • danielgatis rembg prior to 2.0.82

Identifiers

  • CVE-2026-115924
  • GHSA-qr6x-7v33-6qmg

Timeline

  • 2026-10-26: advisory

References

This page as Markdown. Free to reuse under CC BY 4.0 with a link back.