Securing AI Agents with Amazon Bedrock AgentCore

Course Arc

Seven modules that build from threat model to a fully governed production agent. Each module maps to one or more of the seven Amazon Bedrock AgentCore security surfaces.

1 Understand the threat 2 Lock down identity 3 Isolate the runtime 4 Enforce policy

5 Protect memory 6 Observe everything 7 Put it together 8 Prove it live

Hands-on: Security Lab

One live environment that grows with the course. Sign in as a real Amazon Cognito user, chat with a real agent on Amazon Bedrock AgentCore Runtime, then flip a control off and on and watch the same attack leak data or get denied. Every verdict in the Security Trace panel comes from a real AWS authorization decision, not a simulation.

Who Is This For?

Solutions Architects

Designing agent platforms that must pass security and compliance review before production.

Security Engineers

Owning the guardrails, IAM policies, SCPs, and audit posture for agentic workloads.

GenAI Developers

Building and deploying agents on Amazon Bedrock AgentCore who need secure-by-default patterns.

Prerequisites

AWS security fundamentals (IAM, KMS, VPC)
Basic Amazon Bedrock understanding
Familiarity with the concept of AI agents and tool use

Certification alignment: Agentic AI Technical Badge pathway and AWS Security Specialty.

What You'll Be Able to Do

Apply the Shared Responsibility model

Know exactly where AWS's responsibility ends and yours begins — including session-to-user binding.

Scope agent identity

Use Token Vault and the 8 Amazon Bedrock AgentCore IAM condition keys to enforce least privilege per user and session.

Write Cedar policies

Control agent-to-tool access at the Gateway with deterministic, auditable authorization.

Protect memory & data

Enforce CMK encryption, namespace isolation, and memory-poisoning mitigations.

Prove compliance

Trace reasoning end-to-end and validate posture with Security Hub controls.

Compose defense-in-depth

Layer all seven controls into a production-ready agent deployment.

Modules

Seven interactive explainers, one per module. Click any module to open its explainer page.

Hands-on

The live lab the modules build toward — real Cognito, real Amazon Bedrock AgentCore Runtime, real IAM verdicts.

The Seven Security Surfaces

Every Amazon Bedrock AgentCore deployment has these seven surfaces. The course covers each one, grounded in official AWS documentation and AWS blog posts.

Identity

Token Vault, OAuth 2.0, SigV4, API keys, IAM condition keys. M2

Memory

CMK encryption, namespace isolation, poisoning mitigation. M5

Policy (Cedar)

Deterministic, auditable authorization; default-deny. M4

Runtime Isolation

MicroVM per session; control/data plane separation. M3

Network (VPC / SCP)

VPC mode, private connectivity, org-level guardrails. M3

Observability

OTEL tracing, CloudTrail, Security Hub controls. M6

Gateway

Central policy choke point; MCP mediation; interceptors. M4

Source Materials

This course is grounded in official, publicly available AWS documentation and blog posts. Every source is linked below, and each module page cites the specific sources behind its claims.

Public Documentation

AWS Blogs

Note: Every claim in this course is grounded in publicly available AWS material — the documentation and blog posts linked above, plus the Service Authorization Reference for actions and condition keys. Each module page lists its own sources.