Security posture of LLM and Agentic AI
Sponsored By
ArmorCode
Let's question the security posture of Agentic AI and LLMs. Do we even have a way out or a framework for assessing the security posture for these fast-paced, evolving domains
Event Overview
Artificial Intelligence is rapidly evolving from assistants that respond to prompts into autonomous agents capable of reasoning, planning, and interacting with multiple systems.
As organizations adopt Agentic AI, the security landscape is changing just as quickly.
Modern AI applications introduce entirely new attack surfaces through Context Windows, Model Context Protocol (MCP), Retrieval-Augmented Generation (RAG), tool execution, and autonomous decision-making. Securing these systems requires more than traditional application security. It requires rethinking how we design, integrate, and continuously assess AI-powered applications.
Join Threat Modeling Connect (TMC) Toronto Chapter for an evening of practical discussions exploring the evolving world of Agentic AI security and modern threat modeling.
Whether you're already working with AI or just beginning to explore its security implications, this event is designed to provide practical insights, real-world scenarios, and meaningful conversations with the cybersecurity community.
Speakers
Dr Li Yang
Assistant Professor, Ontario Tech University
Talk: "Can We Trust AI to Automate Cybersecurity? From AutoML to LLM-Assisted Autonomous Security"
Vinay Puri
Senior Cybersecurity Advisor at Laurentian Bank
Talk: "The Agent Attack Surface"
Session 1
Speaker: Dr Li Yang
AI is rapidly moving from supporting cybersecurity professionals to taking a more active role in security automation. But should we trust AI agents to make complex ML and security decisions on their own? This talk introduces a practical approach that combines AutoML, multi-objective optimization, and LLMs for autonomous cybersecurity. AutoML automates repetitive model development and optimization tasks, while multi-objective optimization considers real-world trade-offs among detection performance, latency, cost, and computing resources. Rather than allowing LLMs or AI agents to directly control the entire process, we explore how they can assist with understanding requirements, analyzing relevant information, and recommending promising ML development strategies. The talk highlights how these technologies can make ML-based security development more efficient, resource-aware, and reliable.
Session 2
Speaker: Vinay Puri
As AI agents gain access to tools, applications, and browsers, their attack surface grows.
In this session, Vinay Puri will explore how tool access, connected apps through MCP, and browser agents introduce new security risks, including the “confused deputy” problem, where an agent can be manipulated into using its own permissions on an attacker’s behalf.
Through real-world scenarios and collaborative discussion, the session will cover:
How tools, connected apps, and browser agents expand the attack surface
Why prompt injection is difficult to eliminate at the model level
Practical defenses, including least-privilege permissions, human approval for high-risk actions, isolating untrusted content, tool vetting, and activity logging
Attendees will leave with a clear mental model of agent risk and a practical checklist for deploying agents without handing attackers the keys.
Interactive Quiz
Wrap up the evening with a fun Slido Quiz and test your AI security knowledge.
Five winners will receive a complimentary copy of Principle of Building AI Agents.
Agenda
5:30 PM - Doors Open
6:05 PM - Welcome Note
6:20 PM - Session 1: Can We Trust AI to Automate Cybersecurity
6:50 PM - Session 2: The Agent Attack Surface
7:20 PM - Interactive Slido Quiz
7:40 PM - Closing Remarks
7:45 PM - 8:30 PM - Networking & Refreshments
Join the Conversation
The future of AI security is being shaped today. Whether you're designing secure AI systems, modernizing your threat modeling practices, exploring the intersection of AI and cybersecurity, or simply eager to learn from experienced practitioners, we'd love to have you join us.
Register today and be part of an evening of learning, discussion, and networking with Toronto's cybersecurity community.
Venue
Highline Beta
372 Bay St., Suite 200
Toronto, ON M5H 4B2