Analysis
DITTO: A Context-aware Pickle-based Pre-Trained Model Scanner for Effective Security Audits
Publication date unknown · Discovered arxiv.org
Page published
Publication date unknown · First observed: 12 Nov 2026
Coverage timeline
Single-source analysis — one report is available.
Why it matters
Untrusted AI model files can execute code during loading, making artifact provenance and safe deserialization important security controls before deployment.
NeuralTrust’s enterprise guide describes AI model security across model sourcing, training data, stored weights, and production runtime, highlighting code execution when loading untrusted model files. Supporting DITTO research examines Pickle-based model risks and evaluates context-aware scanning against benign and malicious pretrained models.