Safeguard Docs

AI Security Posture Management (AI-SPM)

Check AI model files for hidden code and unsafe deserialization - across Pickle, PyTorch, safetensors, GGUF, and ONNX formats - before untrusted weights load into your environment.

AI Security Posture Management (AI-SPM)

Model weights are executable supply chain artifacts. A malicious pickle can run arbitrary code the moment a model is loaded. AI-SPM inspects model artifacts before they are trusted, running model file and format safety checks on the formats teams actually ship, including unsafe deserialization.

AI-SPM is the lifecycle wrapper around Safeguard's AI security - discover, assess, protect, govern, monitor. Model-artifact checking (below) is the core; it sits alongside AI-BOM discovery, which inventories the models it scans, and the AI Gateway, the runtime LLM firewall.

Model-artifact safety checks

AI-SPM statically analyzes model files without loading them:

FormatWhat AI-SPM checks for
Pickle (.pkl, .bin, PyTorch state dicts)Flags hidden code-execution instructions - imports of dangerous modules like os / subprocess / eval, and load-time steps that can run arbitrary code the moment the file is deserialized. Mapped to CWE-502 (unsafe deserialization) and CWE-94 (code injection).
PyTorch .pt / .zip archivesUnpacks the archive and inspects every embedded component for the same hidden code-execution patterns.
safetensorsConfirms the file is a genuine, tensor-only artifact with no executable payload.
GGUFValidates the container so malformed or tampered files are caught before use.
ONNXInspects the model graph without executing it and flags two risks: (a) custom or non-standard operators and embedded code-execution steps - CWE-94 (code injection); and (b) external-data references whose path escapes the model directory (absolute paths, .., UNC paths, or URLs) - CWE-22 (path traversal). Malformed files fail safe.

Because these formats can execute code the moment a model loads, AI-SPM treats any hidden code-execution instruction as a high-signal finding - the whole point is to catch it before the weights are loaded.

Findings

Each finding identifies the artifact, the format, the specific risk that fired, and its CWE mapping. Findings flow into the unified findings model as AI-security findings, tenant/org-scoped end to end.

Configuration

AI-SPM is admin-toggleable via a feature flag, enforced server-side. Once enabled, model artifacts are scanned as they are ingested through your registries and pipelines.

On this page