AI Security
Protect AI models, applications, and agents.
Browse defensive guidance, tools you can run yourself, and products for protecting deployed AI systems.
AI Security Guides
Plan defenses for AI applications, agents, and model pipelines.
-
AI Data Security
Joint guidance for protecting training, testing, and operational data throughout the AI system lifecycle.
New Zealand National Cyber Security Centre, NSA, and international partners · Public guidance -
Deploying AI Systems Securely
Deployment guidance for organizations operating AI systems, with practical controls for the hosting environment and ongoing maintenance.
National Security Agency and international partners · Public guidance -
Google Secure AI Framework
A catalog of defensive AI controls for securing data, models, infrastructure, and agent permissions.
Google · Public guidance
AI Security Tools
Inspect models, protect sensitive data, and check AI safeguards.
-
Guardrails AI Python Framework
A Python framework for validating model inputs and outputs and enforcing structured responses.
Guardrails AI · Open-source framework -
ModelScan
A pre-load scanner that flags suspicious constructs in supported machine learning model files.
Protect AI · Open-source software -
Safetensors
A format and library for distributing tensor weights without loading Python pickle objects.
Safetensors contributors · Open-source format and library
Commercial AI Security
Compare managed controls for AI applications and model deployments.
-
Amazon Bedrock Guardrails
Managed input and output policies for applications using Amazon Bedrock or the standalone guardrail API.
Amazon Web Services · Paid cloud service -
Cisco AI Defense
Enterprise controls for discovering organizational AI use and protecting AI applications at runtime.
Cisco · Paid enterprise subscription -
Cloudflare AI Security for Apps
Cloudflare WAF controls for discovering AI endpoints and inspecting supported AI requests.
Cloudflare · Free discovery; paid Enterprise detection add-on