Rogue Agents - Stop AI from misusing APIs

Slide 1

Slide 1 text

Rogue Agents Stop AI from misusing APIs TechTalks SF - Jan ‘24 Dominik Kundel d-k.im/rogue-agents Dominik Kundel | @dkundel

Slide 2

Slide 2 text

Dominik Kundel | @dkundel

Slide 3

Slide 3 text

console.log(` Hi! I’m Dominik Kundel `); dkundel.com @dkundel [email protected] github/dkundel Product Lead @ Twilio && JavaScript Hacker Dominik Kundel | @dkundel

Slide 4

Slide 4 text

How can we have AI interact with APIs? Dominik Kundel | @dkundel

Slide 5

Slide 5 text

How can we have AI safely interact with APIs? Dominik Kundel | @dkundel

Slide 6

Slide 6 text

How can we have AI interact with APIs? Dominik Kundel | @dkundel

Slide 7

Slide 7 text

Dominik Kundel | @dkundel How to connect AI to APIs Platforms Frameworks Native LLM Functions 🦜🔗

Slide 8

Slide 8 text

Dominik Kundel | @dkundel Platforms Frameworks Native LLM Functions 🦜🔗 Source: LangChain Documentation

Slide 9

Slide 9 text

Dominik Kundel | @dkundel Platforms Frameworks Native LLM Functions 🦜🔗 Source: LangChain Documentation

Slide 10

Slide 10 text

What’s the problem? Dominik Kundel | @dkundel

Slide 11

Slide 11 text

Dominik Kundel | @dkundel

Slide 12

Slide 12 text

Dominik Kundel | @dkundel

Slide 13

Slide 13 text

Unpredictable Dominik Kundel | @dkundel

Slide 14

Slide 14 text

Dominik Kundel | @dkundel Easily Impressionable

Slide 15

Slide 15 text

Dominik Kundel | @dkundel Source: Simon Willison - Prompt Injections: what’s the worst that can happen?

Slide 16

Slide 16 text

Dominik Kundel | @dkundel Source: Simon Willison - Prompt Injections: what’s the worst that can happen?

Slide 17

Slide 17 text

Dominik Kundel | @dkundel Source: Simon Willison - Prompt Injections: what’s the worst that can happen?

Slide 18

Slide 18 text

Dominik Kundel | @dkundel Rules are “suggestions”

Slide 19

Slide 19 text

Dominik Kundel | @dkundel Source: Simon Willison - Prompt injections explained

Slide 20

Slide 20 text

Dominik Kundel | @dkundel Source: Simon Willison - Prompt injections explained

Slide 21

Slide 21 text

Dominik Kundel | @dkundel The problems with LLMs Unpredictable Easily Impressionable Rules “suggestions”

Slide 22

Slide 22 text

Dominik Kundel | @dkundel

Slide 23

Slide 23 text

How do we “LLM-proof” our APIs? Dominik Kundel | @dkundel

Slide 24

Slide 24 text

Dominik Kundel | @dkundel

Slide 25

Slide 25 text

Dominik Kundel | @dkundel

Slide 26

Slide 26 text

Dominik Kundel | @dkundel

Slide 27

Slide 27 text

Dominik Kundel | @dkundel

Slide 28

Slide 28 text

Dominik Kundel | @dkundel LLM

Slide 29

Slide 29 text

Dominik Kundel | @dkundel

Slide 30

Slide 30 text

Dominik Kundel | @dkundel

Slide 31

Slide 31 text

Dominik Kundel | @dkundel

Slide 32

Slide 32 text

Dominik Kundel | @dkundel

Slide 33

Slide 33 text

Dominik Kundel | @dkundel LLM

Slide 34

Slide 34 text

What security measures? Dominik Kundel | @dkundel

Slide 35

Slide 35 text

Dominik Kundel | @dkundel Security Measures

Slide 36

Slide 36 text

Dominik Kundel | @dkundel Security Measures Data Validation

Slide 37

Slide 37 text

Dominik Kundel | @dkundel Security Measures Data Validation Rate Limiting

Slide 38

Slide 38 text

Dominik Kundel | @dkundel Security Measures Data Validation Authentication Rate Limiting

Slide 39

Slide 39 text

Dominik Kundel | @dkundel Security Measures

Slide 40

Slide 40 text

Dominik Kundel | @dkundel Security Measures Authorization

Slide 41

Slide 41 text

Dominik Kundel | @dkundel Security Measures Authorization Least Privilege

Slide 42

Slide 42 text

Dominik Kundel | @dkundel Security Measures Authorization Eliminate con fi dential & unnecessary data Least Privilege

Slide 43

Slide 43 text

Dominik Kundel | @dkundel Do threat modeling!

Slide 44

Slide 44 text

Dominik Kundel | @dkundel Takeaways?

Slide 45

Slide 45 text

Treat AI-exposed APIs as public Dominik Kundel | @dkundel Takeaways?

Slide 46

Slide 46 text

Treat AI-exposed APIs as public Security mechanisms outside AI world Dominik Kundel | @dkundel Takeaways?

Slide 47

Slide 47 text

Treat AI-exposed APIs as public Security mechanisms outside AI world Dominik Kundel | @dkundel Takeaways? Toddler-proof your home API!

Slide 48

Slide 48 text

console.log(` 💖 Thank You! 🎉 `); dkundel.com @dkundel [email protected] github/dkundel d-k.im/rogue-agents Dominik Kundel | @dkundel | #workfromhomeconf |