The AI stack is becoming increasingly interconnected. Agents now invoke tools, interact with MCP servers, maintain memory, and coordinate with other agents to complete tasks that were once handled by traditional software systems.
This session examines the new attack surface created by agent ecosystems, tool integrations, memory layers, and autonomous workflows. As AI systems become more interconnected and autonomous, trust extends far beyond code and infrastructure, shaping how agents access tools, retain context, make decisions, and interact with one another. The result is a new set of security and governance challenges that traditional software supply chain models were never designed to address.
From agent frameworks to runtime interactions, this talk provides a security-first view of the emerging AI supply chain and the controls needed to govern it.
Roshan Piyush is Head of Security Research at Harness, where he leads research across application, API, runtime, and DevSecOps security. An OSCP-certified security expert, he has spent more than a decade building security products and advancing security engineering at companies including Harness, Traceable, Ping Identity, and Oracle. He is also the lead of the OWASP crAPI project and holds multiple patents in application security and data protection.
