On-demand workshop

OWASP MCP Top 10: Securing the New Trust Boundary for Agentic AI

MCP connects agents to tools, APIs, files, databases, developer environments, cloud services, and enterprise workflows. That power is exciting, but it also creates a new attack surface that many teams are only beginning to understand. In this on-demand session, you'll explore the OWASP MCP Top 10 through practical scenarios, attacker mindsets, and defensive patterns. We will look at risks such as prompt injection through tools and resources, tool poisoning, excessive agency, unsafe tool execution, over-permissioned MCP servers, context leakage, insecure authentication and authorization, supply chain abuse, weak logging, and governance gaps.

Rather than staying theoretical, the workshop will walk through real-world-style examples showing how MCP risks appear in agentic AI workflows and how they can be threat modeled, tested, and mitigated. You'll leave with a clear understanding of MCP security risks, practical review questions, and actionable controls they can apply when building, adopting, or governing MCP-enabled AI systems.

Watch On-Demand

View this session to:

  • Understand why MCP is a new trust boundary for agentic AI.
  • Learn the major security risks covered by the OWASP MCP Top 10.
  • See practical examples of MCP abuse paths and failure modes.
  • Learn how to threat model MCP clients, servers, tools, and resources.
  • Walk away with defensive controls for safer MCP adoption.

Speaker

vandana verma sehgal.png

Vandana Verma Sehgal
Staff AI Security Advocate, Snyk

Snyk is trusted by the world’s most innovative companies.

© 2026 Snyk Limited
Registered in England and Wales
AISecEng Logo