AWS re:Inforce 2025 - COM326

In this AWS re:Inforce 2025 session, Shaoyi Li, a lead cloud engineer at PE Technology and AWS hero, presents an innovative AI-driven solution called "Threat Designer" for automated threat modeling. The presentation focuses on addressing common challenges in threat modeling through a multi-agent system that can analyze system architecture diagrams to identify potential security threats.

The session addresses several key business challenges, including the new risks introduced by AI systems (such as prompt injection and model poisoning), the time-consuming nature of manual threat modeling, and the tendency for security considerations to be deferred until late in the development cycle. The solution integrates both traditional STRIDE framework and new AI-specific threat frameworks like OWASP Top 10 for LLM.

The presented solution, Threat Designer, is a serverless architecture deployed on AWS that automates threat modeling through a five-node workflow system. The tool provides automated threat identification, AWS-based mitigation solutions, and red teaming capabilities through a chatbot interface. The system transforms threat modeling from a static to a dynamic process, enabling real-time updates as systems evolve and promoting a "security by design" mindset within organizations.

Transcript

1. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved.
2. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Hardening Generative AI Applications Using Threat Model Agent Shaoyi Li C O M 3 2 6 (he/him) AWS Community Hero PAX Technology
3. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Agenda 01 Why threat modeling 02 Common challenges with threat modeling 03 How to address the challenges 04 AI-Powered Threat Modeling Solution 05 Architecture overview 06 Demo
4. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Why threat modeling? • Skipping threat modeling leaves applications at high risks (breaches, outages, brand damage). • Shift Left - detect and fix the risks in the design phase can reduce security cost by up to 150x • Meet compliance requirements, such as PCI DSS, GDPR, HIPPA, etc. • A security architecture design best practice highlighted in the AWS Well-Architected Framework (SEC01-BP07 “Identify threats and prioritize mitigations using a threat model”) • Create a security-by-design culture within the organization
5. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Challenges with threat modeling Novel AI Risks Manual and Complex Static and Stale Later Stages in SDLC
6. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. How to address the challenges STRIDE + OWASP Top 10 Automated and Streamlined Continuous Updates Security by Design
7. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. AI-Powered Threat Modeling Solution Threat Designer Threat Designer is an AI-driven agent designed to automate and streamline the threat modeling process for secure system design. Features: •Supports threats identification in AI system •Automated threat modeling from a single architecture diagram •Dynamic threat modeling as the system scales •Mitigate threats in design phase
8. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. AI-Powered Threat Modeling Solution Start Image processing Assets Flows Threats End Continue Standard Replay: skip and retrieve from state Iterations • auto gap analysis • reasoning (4k/8k/16k tokens) Input: • System architecture diagram (required) • Infrastructure as Code template files • System Description & Assumptions Output: • Assets list, data flows, trust boundaries, threat sources • Comprehensive threat list generated from STRIDE & OWASP Top 10 for LLMs • Mitigations, remediation plan and red teaming plan
9. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Architecture overview
10. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Architecture overview • Frontend: Amplify (ReactJS) • Authentication: Cognito • API: API Gateway • Triger Entries: UI and GitHub Push
11. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Architecture overview Backend: Lambda • Operate Threat Models • Create Threat Model with Agent • Chat with Threat Models
12. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Architecture overview Database: DynamoDB • Agent Execution State • Threat Catalog Metadata • Threat Model Chat History
13. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. • AI Model: Claude 3.7 Sonnet on Amazon Bedrock (Converse API) • Object Storage: Pre-signed URL from Amazon S3 • Threats Tracker: Jira REST API Architecture overview
14. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Demo
15. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Demo Upload a architecture diagram Upload an IaC file Enable the OWASP Top 10 for LLM framework Enable gap analysis and reasoning Start threat modeling
16. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Demo Threat modeling by a 5-node agent workflow
17. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Demo Mapped to OWASP LLM Top 10 threat Threat modeling results Clear AWS based mitigation actions
18. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Demo Generate remediation strategies with IaC code Predefined sample prompts
19. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Demo Generate red teaming plan Multiple model selection
20. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Conclusion • AI-Powered threat model agent: automated, streamlined, scalable • Dynamic threat modeling, evolves as the system updates • Aligned with novel AI-specific threats • Security-by-Design mindset: Identify threats in early stage of SDLC. • Conversational mitigation and red teaming solutions Future Plan • Utilize the AI agent to simulate red teaming and patches itself to enhance threat modeling.
21. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Get the Agent on Github! https://github.com/awslabs/threat- designer [Threat Designer: AI-powered threat modeling Agent for secure system design]
22. © 2025, Amazon Web Services, Inc. or its affiliates. All rights reserved. Thank you! Thank you! Please complete the session survey in the mobile app Shaoyi Li in/shaoyi-li-cloud Shaoyi_cloudsec