Breaking Models AI Security

Breaking Models: Session 5 - Leak & Build Edition

Part 1 - LLM02: Sensitive Information Disclosure

Discover the types of sensitive information AI applications can leak, including API keys, customer records, system prompts, and private conversations. This issue ranked #6 on the original OWASP Top 10 for LLMs and has now risen to #2 in the 2026 update, just behind prompt injection. We will explore how these leaks occur, often without the need for jailbreaking, and discuss effective prevention methods.

Part 2 - Design Exercise: Build a Secure Agentic System

In this session, we will not only discuss vulnerabilities but also design a secure agentic AI application. Together, we will threat model potential leak points and hijacking risks, and collaboratively sketch a secure architecture. This is a hands-on design experience, not just a lecture.

Format: Talk, followed by the exercise, and then an open floor for discussion. No vendor pitches. Beginners are welcome!

📩 RSVP or to present at a future session

📅 9 August · 2:30 PM - 4:30 PM

📍 Hacker Dojo, Mountain View, CA 94043, USA