In this week's episode, we're diving deep into the murky world of prompt injection! We explore what it is, how attackers exploit vulnerabilities, and what's at stake if you don't take it seriously.
We discuss real-world examples, from the infamous $1 car deal to the quest for leaked system prompts, highlighting the potential damage of a successful attack. More importantly, we get into practical strategies for defending against prompt injection, including building deterministic guardrails and layering security models.
We break down how to approach security in AI systems, emphasizing that it's just like building good software: layer by layer. Tune in to learn how to protect your AI applications and keep your data safe!
Check out our github: https://www.github.com/boundaryml/baml
AI That Works repo: https://github.com/ai-that-works/ai-that-works
Socials:
X: https://x.com/boundaryml
Discord: https://discord.com/invite/yzaTpQ3tdT
LinkedIn: https://www.linkedin.com/company/boundaryml/
⏱️ Chapters:
Intro
Risks of Automation
More on Prompt Injections
Developer Content & Prompting
Error Handling in Code
AI Engineering is Software
Agent Memory Registers
Network Sniffing Sys Prompts
Wrapping Up
Transcript
I've watched this one closely. Ask me anything about it and I'll point you to the exact moment.