All videos
0:00 / 0:00
ai

How Cogent Security sandboxes AI agents that touch customer data | Max Agency #podcast

LangChain16 June 2026Watch on YouTube

Part of series

Ep. 11 · LangSmith Engine: Agents bouwen

LangChain laat zien hoe LangSmith Engine het evalueren en verbeteren van AI-agents automatiseert via trace-analyse.

View the series

Description

Gng Sng, Co-Founder and CTO of Cogent Security, breaks down how they approach defense in depth for AI agents. Their sandboxes run with read-only permissions by default, a policy engine handles access elevation, and any write action happens outside the box — fully secure and deterministic. The rule: agents should never be able to do something the system is disallowing. This clip is from Max Agency, a #podcast about how the best AI agents are actually being built. Hosted by Harrison Chase, CEO of LangChain, each episode goes deep with the builders designing, deploying, and learning from real agent systems in the wild. From architecture decisions to evals, tooling, and failure modes, Max Agency is for people who want to understand what it really takes to build useful agents. #AIAgents #AISecurity

What you'll learn

  • Cogent Security uses sandboxing to run AI agents with read-only permissions by default, preventing unintended actions
  • A policy engine manages access elevation, ensuring write actions occur outside the sandbox in a fully secure and deterministic manner
  • Defense in depth means agents can never perform actions the system disallows, protecting against misconfiguration or error

Frequently asked questions

How does Cogent Security protect AI agents that handle customer data?
Cogent Security uses sandboxing with read-only default permissions and a policy engine that controls access elevation. Write actions occur outside the sandbox to ensure complete security and determinism.
What is a sandbox and why is it important for production AI agents?
A sandbox is an isolated environment where AI agents run with limited permissions. This prevents agents from performing unauthorized actions on customer data, which is essential for secure production deployment.
What does 'defense in depth' mean in Cogent Security's security model?
Defense in depth means AI agents can never perform actions the system disallows, regardless of misconfiguration or errors. This ensures unexpected situations remain under control.

Topics

ai agentssecuritysandboxproduction deploymentarchitecture