Ensure that your Amazon Bedrock AgentCore gateways have a policy engine configured to evaluate and authorize agent tool calls. A policy engine is a collection of Cedar-based policies that intercepts all agent requests to a gateway and determines whether to allow or deny each action based on the defined policies. Without a policy engine, agents can invoke any tool available through the gateway without additional authorization checks beyond the initial gateway-level authentication.
When a policy engine is associated with a gateway, it provides an additional layer of security by enforcing fine-grained access controls at the tool invocation level. Policy engines use Cedar, a policy language designed for authorization, to define rules that can restrict which tools agents can call, what parameters they can pass, and under what conditions tool calls are permitted. This enables organizations to implement guardrails that prevent agents from performing unauthorized or potentially harmful actions, even if the agent has valid credentials to access the gateway.
Configuring a policy engine on Amazon Bedrock AgentCore gateways is essential for implementing defense-in-depth security for your AI agent infrastructure. While gateway-level authentication ensures that only authorized entities can access the gateway, a policy engine provides granular control over what actions agents can perform once authenticated. This is particularly important for production deployments where AI agents interact with sensitive systems such as databases, customer data, financial systems, or external APIs. Policy engines enable you to implement guardrails that prevent agents from accessing tools they shouldn't use, restrict parameter values to safe ranges, audit and log all policy decisions for compliance, and respond to changing security requirements by updating policies without modifying agent code.
Audit
To determine if your Amazon Bedrock AgentCore gateways have a policy engine configured, perform the following operations:
Remediation / Resolution
To configure a policy engine on your Amazon Bedrock AgentCore gateways, perform the following operations:
Before associating a policy engine with a gateway, you must first create a policy engine and define policies within it. The policy engine uses Cedar policy language for defining authorization rules. For information about Cedar policies, see the Cedar Policy Language Guide.References
- AWS Documentation
- Amazon Bedrock AgentCore Gateway: Securely connect tools and other resources to your Gateway
- Create a policy engine
- Create gateway with Policy Engine
- Update existing gateway with Policy Engine
- Welcome to the Cedar policy language
- AWS Command Line Interface (CLI) Documentation
- list-gateways
- get-gateway
- update-gateway
- create-policy-engine