D@S #56: Building Security for Rabbit's AI-powered Hardware
Matthew Domko explains how his team tackles securing a consumer AI device that can take real-world actions, requiring expertise across hardware, Android, and cloud security.
Detection at Scale is a podcast and a newsletter dedicated to helping security practitioners navigate the complexities of Enterprise-scale detection and response. Each episode features industry leaders who bring real-world experience and actionable lessons to help you stay ahead of the curve!
In this episode of Detection at Scale, Matthew Domko, Head of Security at Rabbit, shares his expertise in building security for AI-powered hardware. Matt's approach combines serverless architecture, infrastructure-as-code, and strategic LLM implementation to create a scalable security operation that spans hardware, Android, application, and cloud domains.
Matt challenges conventional security practices with a data-first approach using AWS Lambda and SQS. His pragmatic philosophy—"I don't see LLMs replacing engineers. I see LLMs making engineers move faster"—drives Rabbit's security operations, where they've discovered that giving AI models more freedom to analyze often leads to better outcomes than overconstraining them.
Topics:
Securing AI-powered hardware that can autonomously interact with third-party services
Achieving 100% infrastructure-as-code coverage while maintaining agility for rapid product iteration
Leveraging serverless architecture (AWS Lambda and SQS) to reduce management overhead and improve scalability
Using private LLMs via Amazon Bedrock to analyze security events and telemetry
Building cost-effective security data lakes with strategic documentation and planning
Transforming security teams from tool operators into engineers with practical implementation of Docker, Lambda, and S3
The counterintuitive approach of giving LLMs more freedom for better analysis rather than overconstraining them
Implementing "detection-as-code" pipelines that leverage infrastructure-as-code for security rules
Watch it on YouTube here:
Chapters:
00:00 Introduction
00:58 Securing AI orchestration systems at scale
04:41 Evolution of spear phishing with LLMs and defensive strategies
09:20 Practical applications of LLMs in detection engineering
13:05 Strategies for secure AI integration in security operations
20:47 Architecting cost-effective security data lakes
23:39 Implementing serverless security architecture with AWS Lambda & SQS
28:31 Infrastructure-as-code approaches for security automation
32:14 Key advice for modern security programs
Last Episode:
D@S #55: Salesforce’s Mor Levi on AI Agents in SecOps
Latest Post:
