Hello AI, Am I Really Safe Here?

Amrit Shenava

Who Am I?

• I am a human being.
• I am an extremely curious and entrepreneurial guy with a builder, hacker and researcher mindset.
• I run a startup called Flashmates which is a full-stack real estate technology platform with over 120k+ monthly active users globally.
• Graduated from Kent State University in Ohio with a Bachelors in Computer Science.

Disclaimer

• This talk is for educational and research purposes only. Misusing your technical skills for a wrong cause can lead to bad outcomes.
• This talk is completely interactive and so please feel free to interrupt me if you have a question or want to add a point.

Purpose Of CyberSecurity

• Computers are helpful but they are stupid.
• They understand 0s and 1s (binary language).
• Humans program computers to perform a task.
• The AI software engineer Devin is also programmed by human software engineers and they are hiring human software engineers to build a AI software engineer.
• We need humans who can detect this stupidity and ensure it is safe from bad actors.

Research Overview

• Accepted and presented at IHCI Uzbekistan 2022
• AI powered IoT devices like Alexa, Nest, Apple HomePod, .etc allow custom apps to be built that can be accessed by interacting with the device.
• If you use your own custom endpoint, you can create your own functions and interact with external API and databases.
• Bad actors can violate and breach privacy or do something dangerous.

Research Explanation

Research Code Snippet

LLM Point Of View

• LLM provides a lot of data and knowledge.
• Most of the apps being built using LLMs are Wrappers around the LLMs.
• LLM Wrappers can be exploited through function calling.
• Function calling is when you want to leverage the LLM to perform a specific task.

LLM Is Tsar Bomba

• Deepfakes
• Voice Impersonation
• Prompt Injection
• Shadow AI

Get In Touch With Me

My WhatsApp/iMessage is +13306763377

Email me at amrit@flashmates.tech