Evidence about AI threats is scattered across incident databases, research, GitHub repositories, model cards, and evaluations.
HuggingThreat gives every AI artifact a shared investigation record. It brings together attributed reports, community audit modules, and scoped results so investigators can examine a public model, share a deliberately backdoored test case, or challenge a reported concern.
Current focus Secret loyalties
Investigate a public Hugging Face model or a deliberately modified research model. Reports, lineage, and completed audits stay attached to the exact artifact.
Use a known backdoored organism as a control, or begin with a concern about an ordinary model.Open the known case →Help extend the network across artifacts, threats, and community audits.
Related checkpoints, separate findings.
Reports linked to testable hypotheses.
New threats and audits beyond secret loyalties.
Comparable evidence for better test selection and future meta-learning.
This audit surfaced behavior that warrants follow-up. It is not proof of a hidden objective.
This module did not surface a signal in this configuration. It does not clear the artifact.
The run completed, but its quality gates did not support an interpretable result.
The run did not execute correctly. No model evidence was produced.
No completed audit run covers this artifact and module.