Guardian Agents

AI agent by Zenity

Secure the Decision with Intent-Aware Guardian Agents

Guardian Agents are specialized security AI agents that provide intent-aware runtime protection for autonomous AI deployments. They monitor agent behavior step-by-step across cloud platforms like AWS and Azure, reconstruct decision chains for investigations, and continuously sharpen security rules to prevent malicious or accidental harm.

Agent specs

The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.

Languages
English

Category: AI Agents

Deployment
Cloud, Homegrown Infrastructure, SaaS-based platform
Security Frameworks
OWASP, MITRE ATLAS
Monitoring Level
Step-level behavioral monitoring
Detection Type
Intent-aware runtime enforcement
Response Modes
Block, modify, or shut down agent
Analysis Scope
Tool calls, memory access, data usage, prompts, commands
  • Intent-aware runtime enforcement for AI agents
  • Reconstructs full agent decision chains for investigations
  • Mapped to OWASP and MITRE ATLAS frameworks
  • Learns from investigations to sharpen future security rules
  • Secures agents on AWS Bedrock, Azure AI Foundry, and Google Vertex AI

Overview

Guardian Agents are specialized security AI agents within the Zenity platform designed to protect autonomous AI deployments in enterprise environments. They provide intent-aware runtime enforcement by monitoring cloud and homegrown agent execution step by step, analyzing tool calls, memory access, and data usage patterns. These agents solve the problem of autonomous agents making unpredictable decisions that bypass traditional security controls, such as chaining unexpected tool calls or misusing credentials. Guardian Agents reconstruct the full decision chain during investigations and continuously learn from every incident to sharpen future security rules. They are purpose-built for agents running on major cloud frameworks including AWS Bedrock, Azure AI Foundry, and Google Vertex AI.

Detection and Response

  • Runtime Monitoring: Guardian Agents monitor cloud agent execution step by step to detect and block unsafe actions in real time.
  • Intent-Aware Detection: The agents analyze tool calls, memory access, and data usage patterns to understand if an agent is behaving consistently with its intended purpose.
  • Decision Chain Reconstruction: Guardian Agents reconstruct the full decision chain when a security incident occurs to identify how an agent reached a specific action.
  • Framework Mapping: The system maps detected threats and agent behaviors to industry-standard OWASP and MITRE ATLAS frameworks.

Governance and Policy

  • Continuous Learning Loop: Guardian Agents learn from every security investigation to sharpen future rules and improve detection accuracy.
  • Policy Sharpening: The agents turn findings from digital forensics and incident response (DFIR) into active policies that catch similar threats in the future.
  • Deterministic Enforcement: The platform provides a foundation of governance that holds at runtime, preventing agents from reasoning their way past soft guardrails.

Deployment and Platforms

  • AWS Integration: Guardian Agents provide specialized security for custom AI agents built on AWS Bedrock and Amazon Bedrock AgentCore.
  • Microsoft Azure Support: The agents secure homegrown AI deployments running on Azure AI Foundry and Microsoft Foundry environments.
  • Google Cloud & Custom Stacks: Security and governance are extended to agents built on Google Vertex AI and an organization's own infrastructure.

What it does

  • Monitor cloud agent execution step by step
  • Analyze agent intent through tool calls and memory access
  • Reconstruct full decision chains for security investigations
  • Block or modify dangerous tool calls before execution
  • Sharpen security rules based on investigation outcomes

Published results

  • 90% reduction in security violations
  • 95% of high-risk violations automatically remediated
  • 90% of existing vulnerabilities remediated within 4 months

Use cases

  • Security teams use Guardian Agents to monitor cloud agent execution step-by-step and block unsafe actions in real time.
  • Incident responders reconstruct the full decision chain of an agent to understand how it reached a specific unauthorized action.
  • Platform owners enforce deterministic policies that prevent homegrown agents from invoking unapproved third-party APIs or tools.
  • Security researchers use the continuous learning loop to turn investigation findings into active policies that catch future zero-click exploits.

Specifications

Integrations
AWS Bedrock, Azure AI Foundry, Google Vertex AI, Microsoft Foundry, Amazon Bedrock AgentCore, OWASP, MITRE ATLAS
Deployment
Cloud, Homegrown Infrastructure, SaaS-based platform
Supported Platforms
AWS Bedrock, Azure AI Foundry, Google Vertex AI, Microsoft Foundry
Security Frameworks
OWASP, MITRE ATLAS
Monitoring Level
Step-level behavioral monitoring
Detection Type
Intent-aware runtime enforcement
Response Modes
Block, modify, or shut down agent
Analysis Scope
Tool calls, memory access, data usage, prompts, commands
Deployment Model
Agentless SaaS platform

How it works

  1. Discovery and Inventory: Zenity's platform discovers and inventories all cloud and homegrown agents across AWS, Azure, and Google Cloud environments.
  2. Continuous Monitoring: Guardian Agents monitor every step of an agent's execution, analyzing tool calls and memory access in real time.
  3. Real-Time Enforcement: The system evaluates actions against intent and policy, blocking or modifying dangerous tool calls before they can cause harm.
  4. Feedback and Optimization: Findings from investigations are used to automatically update and sharpen security policies for future agent actions.

FAQ

What are Guardian Agents?

Guardian Agents are specialized security agents that monitor, enforce, and respond to risks in real time, distinct from the AI agents they are designed to secure.

How do they understand agent intent?

They analyze the full execution path, including prompts, tool calls, commands, memory, and data usage, to determine if an agent's actions align with its intended purpose.

Can they explain how an agent made a decision?

When an incident occurs, the platform reconstructs the entire decision chain, allowing security teams to see exactly how an agent reached a specific action.

Do Guardian Agents improve over time?

Yes, they learn from every investigation and turn those findings into sharper security rules to prevent future occurrences of the same threat.

Added by AntSpace from public sources (zenity.io).

Related

More AI agents by Zenity