Skip to Main Content
Talk Intermediate MIT License

LLM Security in Practice: Real Exploits, Real Defences

Proposal status is Rejected
Session Description

This talk walks through the OWASP Top 10 for LLM Applications with live demos that show real attacks and their mitigations, side by side.

No slides-only theory. Each vulnerability with working exploit:

  • Prompt injection that hijacks a chatbot

  • A poisoned vector database that ranks malicious documents first

  • An AI agent that drops your database

  • Supply chain attacks that exfiltrate secrets via MCP tools

  • And more

7 out of 10 demos use real attacks

Key Takeaways

  • Practical understanding of how LLM applications get exploited

  • Defence patterns: guardrails, output sanitisation, least-privilege tooling, retrieval validation

  • A mental model for threat-modelling AI features before shipping them

References

https://github.com/itsnoobj/llm-owasp-10-attack-mitigate-demo/blob/main/slides.md

Session Categories

Engineering practice - productivity, debugging
Talk License: MIT License

Which track are you applying for?

Security

Speakers

Jeevan D C
Jeevan D C Senior Principal Engineer | Entain

Jeevan D C is a builder and architect who's been automating his way out of boredom since 2015.

He wrote the first line of code for a digital bank that scaled to 1.5 million customers, helped grow the engineering organization from 3 to 300 people, and led a cloud migration with just 4 hours of downtime.

Over the years, he's worked across banking, telecom, insurance, and gaming — designing systems for scale and then fixing the organizational structures around them.

He teaches at HyperVerge Academy, speaks at conferences, and runs a 2k+ member tech community.

These days, he's focused on technology strategy, organizational design, and occasionally breaking LLMs for fun.

https://noobj.me/