Description
Agents become more useful when they can access company knowledge, but permanent access creates unnecessary risk. In this talk, I’ll show how to combine Docker Sandboxes with a Cognee Company Brain to build safer agentic workflows. An agent runs inside an isolated sandbox, receives temporary access to company knowledge only when needed, retrieves relevant context, completes its task, and has that access removed. We’ll explore how this pattern separates execution, memory, and permissions, reducing exposure of credentials and sensitive data while enabling powerful AI workflows. Attendees will leave with a practical architecture to give agents safer access to enterprise knowledge.
____
Mark Lechner is CISO at Docker, where he leads security across the platform millions of developers use to build, ship, and run software - including the agentic workloads reshaping modern engineering. Before Docker, he was CISO at DeepOpinion/Otera driving enterprise AI and LLM agentic security, CISO at Bitpanda scaling security through Europe's largest crypto platform, and Head of Cybersecurity at Solaris Bank. Before that he spent nearly a decade at Morgan Stanley in various security roles including the global leadership of security tooling. His focus today: supply chain integrity, agentic AI security, and secure-by-default developer platforms. He is based in Berlin.--