Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Rogue Agents - Stop AI from misusing APIs
Search
Dominik Kundel
PRO
January 23, 2024
Programming
0
260
Rogue Agents - Stop AI from misusing APIs
Dominik Kundel
PRO
January 23, 2024
Tweet
Share
More Decks by Dominik Kundel
See All by Dominik Kundel
AWS re:Invent '24 - Rogue Agents - Stop AI from misusing APIs
dkundel
PRO
0
62
AI for Marketers Sept '24 - How AI Agents will change your
dkundel
PRO
0
260
AGI Builders July '24 - Rogue Agents - Stop AI from misusing APIs
dkundel
PRO
0
200
AI Engineer World's Fair '24 - Cooking with Fire without
dkundel
PRO
0
210
SIGNAL 2021 - Live Developer Mode
dkundel
PRO
0
180
OpenJS World - What the AST?
dkundel
PRO
0
460
WFHConf - Move to TypeScript at your own Pace
dkundel
PRO
0
330
SFNode '20 - How to move your project to TypeScript
dkundel
PRO
0
340
Node+JS Interactive '19 - When Porgs Scream at Webpack and Other Stories
dkundel
PRO
0
360
Other Decks in Programming
See All in Programming
技術的負債の正体を知って向き合う
irof
0
260
The Past, Present, and Future of Enterprise Java
ivargrimstad
0
490
Cursorハンズオン実践!
eltociear
2
1.2k
kiroとCodexで最高のSpec駆動開発を!!数時間で web3ネイティブなミニゲームを作ってみたよ!
mashharuki
0
920
Pythonに漸進的に型をつける
nealle
1
130
The Past, Present, and Future of Enterprise Java
ivargrimstad
0
650
バッチ処理を「状態の記録」から「事実の記録」へ
panda728
PRO
0
190
Reactive Thinking with Signals and the Resource API
manfredsteyer
PRO
0
110
React Nativeならぬ"Vue Native"が実現するかも?_新世代マルチプラットフォーム開発フレームワークのLynxとLynxのVue.js対応を追ってみよう_Vue Lynx
yut0naga1_fa
2
1.5k
デミカツ切り抜きで面倒くさいことはPythonにやらせよう
aokswork3
0
260
TransformerからMCPまで(現代AIを理解するための羅針盤)
mickey_kubo
7
5.4k
Amazon Verified Permissions実践入門 〜Cedar活用とAppSync導入事例/Practical Introduction to Amazon Verified Permissions
fossamagna
2
100
Featured
See All Featured
4 Signs Your Business is Dying
shpigford
185
22k
YesSQL, Process and Tooling at Scale
rocio
173
15k
Large-scale JavaScript Application Architecture
addyosmani
514
110k
The Pragmatic Product Professional
lauravandoore
36
7k
Visualization
eitanlees
149
16k
Unsuck your backbone
ammeep
671
58k
The Psychology of Web Performance [Beyond Tellerrand 2023]
tammyeverts
49
3.1k
Documentation Writing (for coders)
carmenintech
75
5.1k
VelocityConf: Rendering Performance Case Studies
addyosmani
333
24k
Speed Design
sergeychernyshev
32
1.2k
Evolution of real-time – Irina Nazarova, EuRuKo, 2024
irinanazarova
9
1k
Optimising Largest Contentful Paint
csswizardry
37
3.5k
Transcript
Rogue Agents Stop AI from misusing APIs TechTalks SF -
Jan ‘24 Dominik Kundel d-k.im/rogue-agents Dominik Kundel | @dkundel
Dominik Kundel | @dkundel
console.log(` Hi! I’m Dominik Kundel `); dkundel.com @dkundel
[email protected]
github/dkundel
Product Lead @ Twilio && JavaScript Hacker Dominik Kundel | @dkundel
How can we have AI interact with APIs? Dominik Kundel
| @dkundel
How can we have AI safely interact with APIs? Dominik
Kundel | @dkundel
How can we have AI interact with APIs? Dominik Kundel
| @dkundel
Dominik Kundel | @dkundel How to connect AI to APIs
Platforms Frameworks Native LLM Functions 🦜🔗
Dominik Kundel | @dkundel Platforms Frameworks Native LLM Functions 🦜🔗
Source: LangChain Documentation
Dominik Kundel | @dkundel Platforms Frameworks Native LLM Functions 🦜🔗
Source: LangChain Documentation
What’s the problem? Dominik Kundel | @dkundel
Dominik Kundel | @dkundel
Dominik Kundel | @dkundel
Unpredictable Dominik Kundel | @dkundel
Dominik Kundel | @dkundel Easily Impressionable
Dominik Kundel | @dkundel Source: Simon Willison - Prompt Injections:
what’s the worst that can happen?
Dominik Kundel | @dkundel Source: Simon Willison - Prompt Injections:
what’s the worst that can happen?
Dominik Kundel | @dkundel Source: Simon Willison - Prompt Injections:
what’s the worst that can happen?
Dominik Kundel | @dkundel Rules are “suggestions”
Dominik Kundel | @dkundel Source: Simon Willison - Prompt injections
explained
Dominik Kundel | @dkundel Source: Simon Willison - Prompt injections
explained
Dominik Kundel | @dkundel The problems with LLMs Unpredictable Easily
Impressionable Rules “suggestions”
Dominik Kundel | @dkundel
How do we “LLM-proof” our APIs? Dominik Kundel | @dkundel
Dominik Kundel | @dkundel
Dominik Kundel | @dkundel
Dominik Kundel | @dkundel
Dominik Kundel | @dkundel
Dominik Kundel | @dkundel LLM
Dominik Kundel | @dkundel
Dominik Kundel | @dkundel
Dominik Kundel | @dkundel
Dominik Kundel | @dkundel
Dominik Kundel | @dkundel LLM
What security measures? Dominik Kundel | @dkundel
Dominik Kundel | @dkundel Security Measures
Dominik Kundel | @dkundel Security Measures Data Validation
Dominik Kundel | @dkundel Security Measures Data Validation Rate Limiting
Dominik Kundel | @dkundel Security Measures Data Validation Authentication Rate
Limiting
Dominik Kundel | @dkundel Security Measures
Dominik Kundel | @dkundel Security Measures Authorization
Dominik Kundel | @dkundel Security Measures Authorization Least Privilege
Dominik Kundel | @dkundel Security Measures Authorization Eliminate con fi
dential & unnecessary data Least Privilege
Dominik Kundel | @dkundel Do threat modeling!
Dominik Kundel | @dkundel Takeaways?
Treat AI-exposed APIs as public Dominik Kundel | @dkundel Takeaways?
Treat AI-exposed APIs as public Security mechanisms outside AI world
Dominik Kundel | @dkundel Takeaways?
Treat AI-exposed APIs as public Security mechanisms outside AI world
Dominik Kundel | @dkundel Takeaways? Toddler-proof your home API!
console.log(` 💖 Thank You! 🎉 `); dkundel.com @dkundel
[email protected]
github/dkundel
d-k.im/rogue-agents Dominik Kundel | @dkundel | #workfromhomeconf |