Tirsdag – 1 · Frimurerlosjen, rom 1
Solutions Engineer, Solo.io
Steven Thwaites is a Principal Solutions Engineer at solo.io, where he stays close to innovation by working with customers in the cloud-native ecosystem. Working with containers and Kubernetes since 2016 — at Docker and AWS — Steven helps developers build, share, and operate their software through robust, open-source solutions. He now is focused on the practical application of AI and overcoming hurdles to its safe introduction into the Enterprise.
Your AI agents have trust issues. And they should. They're talking to external LLMs, sharing your proprietary data, and making autonomous decisions that could expose your business to prompt injection attacks, data leakage, and runaway costs. Traditional API gateways and WAFs can't help—they're blind to the "Layer 8" conversation happening inside the HTTP body.
In this talk, we'll explore why securing AI applications demands a fundamentally different approach than securing traditional microservices. We'll dissect the unique attack vectors facing LLM-based applications—from jailbreaking and prompt injection to data exfiltration via tool misuse—and demonstrate how to build a defense-in-depth strategy.
We'll introduce the concept of Model Context Protocol (MCP) awareness at the gateway level, showing how to inspect, filter, and govern AI traffic without breaking the application flow.
You'll see a live demo of:
Real-time Prompt Injection Detection: Blocking malicious inputs before they reach the model.
Data Loss Prevention (DLP): Redacting sensitive PII from model responses on the fly.
Agent Identity & Authorization: Enforcing strict policies on which agents can access which tools and data.
Leave this session with a concrete blueprint for securing your AI agents in production using open-source tools, moving from "move fast and break things" to "move fast and stay secure."
Programmet er oppdatert.