ARS Session 05 - Likhitha Lakshmi Gudivada

Security Risks and Guardrails in Large Language Models: A Practical Analysis of LLM Vulnerabilities and Safety Mechanisms Description:This project focuses on improving the security, safety, and reliability of Large Language Models (LLMs) by implementing guardrail-based protection mechanisms. The project examines major LLM vulnerabilities such as prompt injection attacks, hallucinations, unsafe content generation, and sensitive data leakage, which can impact the trustworthiness of AI systems in real-world applications. A secure LLM workflow prototype was developed using AWS cloud services including Amazon API Gateway, AWS Lambda, Amazon Bedrock, Bedrock Guardrails, AWS CloudWatch, and AWS CDK. The system performs input validation, content filtering, risk classification, output moderation, and logging/monitoring to detect and reduce unsafe interactions before responses are returned to users. The overall goal is to demonstrate practical security mechanisms that can support safer deployment of AI-powered applications in cloud environments.