| Framework Home | Research Index | Cross-Pillar Governance | AISM | NEXUS | Dashboard |
Current framework context: AI SAFE² v3.1. This research note preserves its original publication date, evidence, and historical framework references. Use current v3.1 normative control and profile documents for implementation or conformance decisions.
ID: RN-2025-004 | Related Control: [P1.T1.2_ADV], [P2.T2.3_ADV] | Status: Verified
Model Pickle Attacks: Deserialization vulnerabilities in standard model formats (Pickle, PyTorch) allow attackers to execute arbitrary code (RCE) simply by loading a model file.
model.load(), the server is compromised.We move from “Implicit Trust” (downloading from the internet) to “Cryptographic Verification.”
Implementation of the Sigstore infrastructure. Before any model weights are loaded into GPU memory, the system verifies the cryptographic signature against the organization’s trusted root.
A centralized ledger mapping every deployed model to its SHA-256 hash. Any deviation in hash value during runtime triggers an immediate lockdown (Pillar 3).
| Previous research note | Research Index | Next research note |
| Framework Home | Cross-Pillar Governance | NEXUS | Challenge Lab |
AI SAFE² v3.1 | Cyber Strategy Institute