Guardian

Zero Trust for ML Models

Enable enterprise level scanning, enforcement, and management of model security to block unsafe models from being used in your environment, and keep your ML supply chain secure.

Screenshot of Guardian product dashboard

Stay Secure Without Disrupting AI Innovation

The use of ML Models without security checks puts your enterprise at risk. ML Models can contain malicious code and then be downloaded and deployed in your environment, with no policy or controls in place to scan for hidden risks that can be exploited by internal and external cybersecurity criminals.

Guardian continuously scans both third and first party models for security threats, before the model is utilized. This adds a critical layer of security to ML model development and deployment, so you can continue AI exploration and innovation with confidence.

Enforce Model Security

AppSec teams can protect against threats with enterprise wide policies to enforce the use of secure models within the organization and improve security posture.

Ease of Integration

It integrates easily with existing MLOps workflows, to keep pace with AI driven innovation.

Comprehensive Model Scanning

Guardian goes beyond traditional malware scanning tools and detects malicious code in ML models early in development and deployment.

Secure Your ML Models From Unique Threats

Enterprise-Grade Security Capabilities

Guardian ensures only secure and compliant models are utilized in enterprise networks with secure gateways and configurable policies.

Innovative Model Scanning

Guardian’s proprietary vulnerability scanners detect malicious code in new and existing AI/ML artifacts saved in popular ML storage formats.

Aggregated Insights

Guardian includes executive level dashboards that aggregate information on what open source models are being used, if they are secure, who created them, how they are licensed, and more.

Data Integration

Easily integrates with Protect AI’s end-to-end visibility product, Radar, to understand all AI/ML threat surfaces in your organization.

Key Features

Two overlapping screenshots of a block of code and a prompt to scan a first party model

Screenshot of Guardian product showing an overview of first and third party models.

Two overlapping screenshots of the Scan Log in the Guardian product navigation and the product dashboard showing two circle graphs with data on total scans and failed scans.

Secure Gateway

Secure your ML model supply chain with Guardian’s endpoint to capture and check the MLOps team’s model requests before delivery.
Supports Leading Model Formats

Guardian supports Pytorch, Tensorflow, XGBoost, Keras, and other popular frameworks.
Policy Engine

Guardian comes with out of the box policies for model acceptance or rejection, including the enforcement of checks for where the model is from, malicious code, safe storage formats, OSS licensing and more.
Automated Scanning

Embed Guardian's Scanner API in your development and deployment pipelines to scan models before they are distributed, ensuring they are safe to use.

Guardian vs PAI Open Source Comparison

Feature	Guardian	PAI Open Source
Detects serialization attacks in model formats	17	9
Detects serialization embedded in model layers
Detects model architectural backdoors in file formats without serialization risks
Block access to models that don’t comply with policies
SDK support for easy integration in CI/CD or model pipelines
Policy management for all models
Integration into model registries