Gated security proof-of-concept
This repository is publicly accessible, but you have to accept the conditions to access its files and content.
This repository contains a coordinated-disclosure security proof-of-concept for a trust_remote_code consent issue in the GLiNER model loader. It contains no user data and no network callback beyond normal Hub/model load. Access is restricted to the affected maintainers and the huntr / ProtectAI triage team for verification and remediation only.
Log in or Sign Up to review the conditions and access this model content.
GLiNER loader — forced trust_remote_code consent-bypass PoC (gated)
This is a security research proof-of-concept for the huntr Model File Formats program
(format json). It demonstrates that stock GLiNER.from_pretrained can execute attacker
code from a model package without any user trust_remote_code opt-in.
The payload is a minimal GLiNER-shaped model directory used only to show the consent defect. There is no weaponization beyond a local marker file, no user data, and no outbound exfil.
- Class: consent-bypass / unsafe remote code load (CWE-94 / CWE-829)
- Affected: GLiNER (PyPI
gliner, urchade/GLiNER) encoder/decoder HF load path - Disclosure: coordinated via huntr MFF
Full technical details and regenerator are in the gated files / huntr report.
Contact: security researcher mrw0r57 (huntr: ssjcorpsec).
- Downloads last month
- -