Dive into the wild world of prompt injection attacks in LLM-powered web apps! As AI chatbots and assistants become ubiquitous, a new breed of security vulnerabilities emerges. In this hands-on adventure, we'll dissect vulnerable chatbots, craft sneaky exploits, and explore robust defense strategies.

Get ready to break (and fix) things as we create a simple LLM-powered app, then systematically exploit its weaknesses. You'll learn to identify common vulnerabilities, understand the anatomy of prompt injection attacks, and implement effective countermeasures. Perfect for devs who want to build safer AI-driven interfaces and stay ahead in the AI security game!

This talk is designed to be adaptable to different time slots (30-50 minutes) and audience levels. I can adjust the depth of technical content and number of live demonstrations based on the conference schedule and attendee profile.

I plan to provide a GitHub repository with code samples and resources for attendees to reference after the talk. This will include examples of vulnerable chatbots and secure implementations, allowing developers to continue learning and experimenting on their own.

The presentation will include interactive elements where audience members can participate in identifying vulnerabilities and suggesting mitigation strategies. This approach helps reinforce learning and keeps engagement high throughout the session.

While the talk focuses on front-end development, the principles discussed are relevant to anyone working with LLMs in web applications. I'm happy to tailor the content to emphasize aspects most relevant to your conference's focus areas.

I'm open to participating in panel discussions or Q&A sessions related to AI security if that aligns with your conference format. My background in both frontend development and AI security could bring a unique perspective to such discussions.