Speak at The Fifth Elephant 2026 Annual Conference

{Describe your session in 2 paragraphs}

We’re building AI agent architectures that look a lot like distributed systems, passing data across multi-agent pipelines, MCP tools, and external APIs. But because these workflows run autonomously and at incredible speeds, they introduce a brand-new challenge to our standard network models. Once an agent is given tool access to do its job, it executes sub-tasks in the background where it’s tough to get real-time visibility. Things can move sideways fast: agents can chain together perfectly valid individual actions in completely unexpected ways, accidentally move data across boundaries, or call external services in ways a human couldn’t predict. We have great access controls for who gets permission to start a process, but once the agent workflow takes off, tracking exactly how data flows across those fast-moving boundaries becomes a massive blind spot.

In this talk, I’ll walk you through a live, realistic multi-agent pipeline to show you exactly how this visibility gap plays out—even when every single step looks completely green on paper. We’ll look at real-world attack vectors like EchoLeak, where an agent is tricked by ambient data (like a malicious incoming email) into quietly shipping private cloud data to an outside server. Then, I’ll introduce Agent Mesh, a lightweight network layer for agentic systems inspired by traditional service mesh patterns. It sits right in the execution path, treating the agent runtime as an unverified network actor and intercepting outbound requests to enforce simple, clear policy rules. Using a working prototype, I’ll show how we can get data flow control, execution tracing, and blast-radius isolation across agents and MCP-style tools without touching a single line of your agent logic. The goal is to bring the cloud-native infrastructure patterns we already know and trust into the AI space, making these highly autonomous systems observable, secure, and ready for production.

{Mention 1-2 takeaways from your session}

1. Why agent systems need a zero-trust model, similar to microservices—because access control alone does not constrain how agents behave at runtime
2. How to apply service mesh-style enforcement (interception, policy, observability) to control agent-to-tool and agent-to-agent interactions in production systems

{Which audiences is your session going to beneficial for?}
• Engineers building agent pipelines and tool integrations
• Platform / infra engineers working with service mesh, APIs, or distributed systems
• Developers experimenting with multi-agent systems (LangChain, AutoGen, MCP ecosystems)
• Anyone trying to observe or control what agents actually do in production

{Add your bio - who you are; where you work}
Kanika Singhal is a technical leader in Flow Network Security at Nutanix with over 15 years of experience building distributed systems, networking, and security infrastructure. She has spent much of her career designing and implementing production-grade systems and continues to work hands-on with code and system design. Prior to Nutanix, she worked as a developer on SD‑WAN technologies at VMware.

{Add the link to draft slides - PDF/PPT - with comments access}
https://drive.google.com/file/d/1h4SZ5q0PHBukdWykY3zSf6F-g-nna8I2/view?usp=sharing

{Add the link to 2-min elevator pitch video}