LLM Scanning

A threat can be baked into the model artifact before it ships, or it can live in how the model responds once deployed. LLM Scanner covers both, through two complementary detection capabilities.

Feature	Description
LLM Static Analysis	Malicious payloads and arbitrary code-execution paths can hide inside model files. The scanner detects the correct format, then parses it down to its structure to surface embedded payloads, dangerous imports, and anomalous signatures.
LLM Dynamic Analysis	Even a clean model file is no guarantee of safe behavior. The scanner evaluates the deployed model's behaviors itself, dynamically probing it with adversarial prompts to expose prompt injection, sensitive-data disclosure, and other weaknesses that only appear at inference time.

Last updated on

Was this page helpful?