Explore a practical framework to think about and build secure AI agents on Amazon Bedrock AgentCore. This session covers threat modeling specific to agentic workloads, including how agents interact with tools, memory, and external systems, and what you need to watch out for. You'll learn how to apply AWS security best practices across AgentCore services, and walk away with actionable patterns. Suitable for developers and architects building AI agents who want to move from prototype to production with confidence.
What this session is about
Playbook
Editorial commentary · what to actually do about this on Monday
The concept
Threat-modeling agentic workloads specifically — tool surface, memory, external system interactions. Apply AWS security best practices across AgentCore components.
Why it matters
Agent threat models differ meaningfully from API threat models. Your existing security playbook misses tool injection, memory poisoning, and prompt-via-data attacks.
The hard parts
Tool injection (an agent tricked into calling a tool with malicious args). Memory poisoning (an agent's long-term memory corrupted by an attacker). Prompt-via-data (instructions embedded in retrieved content).
Playbook moves
(1) Run STRIDE per agent component (memory, tool registry, planner, executor). (2) Specifically check: who can mutate memory, who can register new tools, who validates tool outputs before they're trusted. (3) Sandbox every tool execution that touches external systems.
The surprise
The most common agent vulnerability in real deployments is *over-broad tool scopes*. Engineers grant tools more permission than needed because it's faster. Audit tool permissions like you audit IAM. Most don't, and it shows up as blast radius when things go wrong.
---
Independent editorial perspective — not an official AWS or speaker statement. Designed for executives evaluating what to brief their teams on next.
Live updates related to this session LIVE
Sourced via Parallel AI Monitor — continuous web watch on 21 topical streams. Updated .
- nist.gov high confidence AI agent regulation & policy
AI Agent Standards Initiative | NIST
OWASP released the 'Top 10 for Agentic Applications 2026', establishing the first dedicated security framework for autonomous AI agents to address vulnerabilities and risks associated with agentic AI systems.
- azion.com Scaling infra for agent workloads
Understanding Agentic AI Infrastructure
Gravitee published an update explaining that the stateless Model Context Protocol (MCP) specification simplifies horizontal scaling for agent workloads. The update also highlights AI gateway governance as a way to manage agent tool traffic, security, and scaling across MCP-based
- newsroom.workday.com high confidence Agent dev tools & observability
Workday Launches New Tools for Developers to Build, ...
AWS launched several Bedrock AgentCore updates on June 17, 2026, including the Optimization Loop for monitoring production traces (failure, intent, and trajectory insights) and the general availability of the AgentCore harness for building and orchestrating agents. Pricing is int
- github.com Agent memory & RAG architectures
I think the Agent should be able to learn from experience ...
Researchers published a study of the gap between retrieving memories and using them faithfully in agent behavior. The work introduces a context-grounded evaluation of agentic memory and examines whether retrieved memories are actually supported by the current context, addressing
- developers.openai.com Scaling infra for agent workloads
Production best practices | OpenAI API
Microsoft’s Azure Foundry Agent Service limits documentation was updated with current guidance that model-call rate limiting is applied at the model-deployment level, with Azure OpenAI quotas and limits determining model-specific constraints. This provides an explicit concurrency
External links matched to this session via topic relevance. The KB does not endorse third-party content; verify before citing.