Securing RAG Applications at AWS Community Day Bangladesh 2025
A conference session on practical security risks in RAG systems, including prompt-injection vectors, pipeline vulnerabilities, and production-grade mitigation strategies.
Topic: Threats, mitigations, and strategy patterns for secure RAG deployments
Target audience: AI Engineers, Solution Architects, Platform Security Teams
Location: BRAC University, Dhaka
Related stack: AWS, Amazon Bedrock, RAG, GenAI Security
Event link: https://www.linkedin.com/feed/update/urn:li:activity:7406057695193157633/
Photo gallery
Session overview
This talk at AWS Community Day Bangladesh 2025 explored how Retrieval-Augmented Generation (RAG) systems introduce a wider attack surface than traditional application stacks, especially when retrieval, orchestration, and prompt assembly are combined in production workflows.
Topics covered
- RAG architecture fundamentals and trust boundaries.
- Attack vectors across ingestion, retrieval, and generation stages.
- Direct and indirect prompt-injection risks.
- Common implementation pitfalls in AI-enabled products.
- Workflow-level vulnerabilities and abuse scenarios.
- Mitigation strategies for production-grade deployments.
- Architecture review patterns for secure GenAI delivery.
Key takeaways
- Security controls must be placed across the full RAG pipeline, not only at model invocation.
- Prompt hygiene and content provenance checks are essential for reducing injection risk.
- Threat modeling should be part of early architecture design, not a post-launch activity.
- Team readiness improves when AI engineering and security engineering collaborate on shared guardrails.