Runtime governance for AI agents with Noma Security and AgentCore Gateway
AWS Partner Network Blog
This article describes how Noma Security's AI Detection and Response (AI-DR) integration with Amazon Bedrock AgentCore Gateway provides runtime governance for AI agents through semantic evaluation of content and intent.
- Runtime governance assesses whether proposed agent actions match user intent, complementing IAM permissions and content controls
- Three core questions: Can content be trusted? Is the actor entitled? Does the action match user intent?
- Noma inspects tool calls, HTTP traffic, and model prompts/responses across MCP, HTTP, and inference targets
- Deployment options include Noma-managed service or customer-managed broker on Amazon EKS
- Shared session history connects requests, responses, and policy decisions for security investigation
- Example: Blocks unauthorized payment updates in procurement tasks despite agent permissions
- Complements Amazon Bedrock AgentCore Policy and Bedrock Guardrails with semantic evaluation
Runtime governance enables secure agent adoption by detecting untrusted instructions and ensuring proposed actions align with user tasks without adding security code to each agent.
The AWS News Feed is currently looking for gold sponsors. If you want to support the AWS community and reach a large audience of AWS professionals, consider sponsoring the AWS News Feed.
Related articles
2026
2026
2026
2026
The AWS News Feed is currently looking for silver sponsors. If you want to support the AWS community and reach a large audience of AWS professionals, consider sponsoring the AWS News Feed.