Is CVE-2026-108943 exploited in the wild?

No. Junglewise Threat Intelligence has no report of CVE-2026-108943 being exploited in the wild.

How severe is CVE-2026-108943?

Medium severity, with a CVSS score of 5.3, published 13 October 2026.

How is CVE-2026-108943 attacked?

It is attacked with local access, in a software library.

Am I affected by CVE-2026-108943?

Check whether you run ModelTC LightX2V 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.

CVE-2026-108943: ModelTC LightX2V code injection in tokenizer loader

CVE-2026-108943 · Severity: medium · CVSS 5.3 · Published 2026-10-13

LightX2V is a framework for generating images and videos from text or other inputs. The framework unconditionally executes arbitrary Python code from model bundles when loading a tokenizer, allowing an attacker who can supply a crafted model bundle to run malicious code with the privileges of the LightX2V process. This vulnerability affects users who load untrusted or third-party model bundles.

Vendors: ModelTC.

Quick answers

Is CVE-2026-108943 exploited in the wild?: No. Junglewise Threat Intelligence has no report of CVE-2026-108943 being exploited in the wild.
How severe is CVE-2026-108943?: Medium severity, with a CVSS score of 5.3, published 13 October 2026.
Which products and versions are affected?: ModelTC LightX2V up to 0.5.0.
Is there a fix for CVE-2026-108943?: No fixed version is recorded yet. Do not load model bundles from untrusted or unverified sources. Restrict write access to model directories and use process isolation (containers or sandboxes) when processing user-supplied model bundles. 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-2026-108943 attacked?: It is attacked with local access, in a software library.
Am I affected by CVE-2026-108943?: Check whether you run ModelTC LightX2V 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

Do not load model bundles from untrusted or unverified sources. Restrict write access to model directories and use process isolation (containers or sandboxes) when processing user-supplied model bundles.

Technical details

The vulnerability is a code injection flaw where AutoTokenizer.from_pretrained() is called with a hard-coded trust_remote_code=True parameter on user-supplied model bundle paths in the Cosmos3 runner. An attacker can craft a model bundle with a malicious auto_map entry in tokenizer_config.json that executes arbitrary Python during tokenizer loading. Additionally, safetensors metadata can override the selected runner, routing execution through the vulnerable Cosmos3 path. The attack requires local access and a victim to load an attacker-controlled bundle; there is no fix published yet.

Affected products

ModelTC LightX2V up to 0.5.0

Timeline

2026-09-24: disclosed
2026-10-13: advisory

References

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