CVE-2026-22584
Uni2TS model configuration permits code injection
Salesforce · Uni2TS
First documented disclosure · Review checkpoint
Summary
Model metadata can reach executable configuration handling when vulnerable Uni2TS loads a model.
Severity and exploitation evidence
High · numeric CVSS score not recorded. Severity source.
Exploitation not established: The reviewed sources establish a disclosed weakness or researcher demonstration, not malicious exploitation in the wild.
Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.
How the weakness works
Hydra instantiation processes attacker-controlled model configuration.
Conditions required
Loading a crafted model with the affected library.
Potential impact
Code execution with the model-loading process’s authority.
Affected versions and configurations
Uni2TS through 1.2.0 on macOS, Windows and Linux, according to the CNA.
Fix and mitigation status
Unit 42 reports an allowlist/validation fix on July 31, 2025; a numbered fixed release was not established from the accessible sources.
Practical next steps
Verify the upstream fix is present before loading third-party models; isolate model-loading jobs.
Operating lesson
A safe weights format does not make the surrounding model loader safe.
Evidence and limits
January 9 CNA publication precedes January 13 technical coverage. The Salesforce help page did not expose substantive text; fix timing is attributed to Unit 42, not independently read from that page. The July 2025 remediation predates this rolling window; January 9 is the earliest public security disclosure date verified in this review, not the date the code was first fixed.
Sources
- Salesforce CNA record
Primary source ·
- Unit 42 AI/ML library research
Primary source ·