Edge Security in the Age of AI
As AI-powered systems move from experimentation into core production workflows, the edge is rapidly becoming the most complex and least understood part of modern architecture. It’s where traffic, identity, policy, and now autonomous decision-making intersect; often in ways traditional systems were never designed to handle.
This meetup brings together Kong and Harness to examine what actually changes when AI moves from demo environments into production systems and what it takes to secure, govern, and operate those systems without slowing teams down.
Kong sits in the critical path of API traffic and service connectivity, enforcing identity, policy, and runtime controls across distributed environments.
Harness is the AI Software Delivery Platform for everything after code embedding intelligence across testing, deployment, security, governance, and optimisation to help teams ship systems faster, safer, and at scale.
Together, we’ll unpack how AI workloads are reshaping the edge. Not in theory, but in the messy reality of production: autonomous agents invoking tools, dynamic reasoning chains, delegated permissions, and failure modes that don’t look like traditional outages.
This session is built for engineers who are actively shipping, securing, and operating AI-enabled systems and want clearer mental models for what’s breaking, why it’s breaking, and how to design for it.
Rather than abstract frameworks, expect grounded discussion, practical lessons, and patterns that can be applied immediately inside real delivery pipelines.
Date: 28 February 2026, Saturday
Location: Harness.io R&D Labs, Bengaluru
Maps: Harness Office
Agenda
10:00 – 10:25 AM
Introductions: Kong × Harness collaboration overview
10:30 – 11:00 AM
Topic: From Requests to Reasoning: Rethinking Edge Defence
Speaker: Roshan Piyush | Head of Security Research @ Harness
Edge security evolved in a world where users, sessions, and request-response cycles were relatively well understood. APIs added scale and statelessness but still operated within predictable trust boundaries.
AI changes that.
When reasoning models and agents enter the request path, the edge is no longer protecting a single transaction. It is governing multi-step execution, tool invocation, delegated authority, and context propagation across systems. A single prompt can influence downstream integrations, trigger chained actions, or escalate privileges in subtle ways.
In this session, Roshan explores how AI reshapes threat models at the edge from identity propagation and tool-level access control to reasoning manipulation and privilege amplification. Traditional request-centric controls are no longer sufficient in an agentic architecture.
Takeaway: A practical mental model for designing edge defence when AI systems are making decisions.
11:10 – 11:40 AM
Topic: Governing Centralised AI & MCP Access for Agentic Systems
Speaker: Suprit Kulkarni | Senior LLM Platform Engineer @ Kong
As organisations move from chatbot experiments to agentic systems, governance becomes the constraint. Without structured control, agents can over-execute, access unintended systems, or drive runaway costs.
This session explores how centralised governance through API gateways and model access control brings visibility, policy enforcement, and cost guardrails to AI-native workflows. The focus is on decoupling agent logic from governance so teams can move fast without compromising safety.
Takeaway: How to introduce strong runtime governance for AI systems without turning guardrails into bottlenecks.
11:50 – 12:40 PM
Panel Discussion: Operating AI at the Edge : Identity, Failure & Incident Response
Featuring Kong + Harness + Industry SRE perspectives
Speakers: TBA
When AI becomes part of the production request path, the “edge” stops being a network boundary and starts becoming a decision boundary.
In agentic systems, actions may flow from a human prompt to a model, to a tool, to downstream services, all within seconds. That raises uncomfortable but necessary questions: Who is the real actor during execution? What does least privilege look like when an LLM can dynamically choose which tools to invoke? And where does accountability sit when something goes wrong?
This panel will explore how identity, guardrails, and incident response must evolve in AI-native architectures. Are traditional SIEM/SOC pipelines enough? How do you reconstruct a reasoning chain after a bad outcome? And what new failure modes are SRE teams already seeing in AI-driven workflows?
Expect a grounded conversation on the trade-offs between protection and velocity and what it actually takes to operate AI systems safely at the edge in 2026.
Takeaway: Concrete patterns for securing, observing, and responding to AI-driven failures, without compromising speed.
12:30 PM onwards
Networking & Hallway Conversations
Coffee + Lunch
Community Connect
Who Should Attend
Site Reliability Engineers (SREs)
Platform & Infrastructure Engineers
Security & Edge Engineers
API Gateway & Traffic Engineers
Engineers building or operating AI-powered systems
Community
This is an engineer-first, community-driven meetup focused on real systems, real failure modes, and real production lessons.
Follow Harness on LinkedIn: Harness
Follow Kong on LinkedIn: Kong