Session: 2 for 1: Plotting a Path to Opening and Controlling the Boundary Between Agents and Real World Systems / Fool AI Once, But Not Twice
Plotting a Path to Opening and Controlling the Boundary Between Agents and Real World Systems – Craig McLuckie
MCP started as a protocol to formalize the interface between stochastics systems (LLMs) and the unforgiving world of deterministic systems (APIs). MCP is seeing breakout engagement and adoption across the industry because it creates a way to make transformers more useful by giving them access to real world data and real world tools. In this session we will explore the role that MCP plays in agentic workflows and the critical considerations when building servers to provide optimal results. This session looks at practices to drive the performance and effectiveness of agents while balancing very real concerns around security, observability and privacy.
Fool AI Once, But Not Twice – Brinnae Bent
In this engaging session, participants will discover how easily AI models can be fooled, misled, or manipulated through various techniques including prompt engineering, adversarial attacks, and data poisoning. Through compelling examples, attendees will gain insights into the fundamental limitations of even the most advanced AI systems and why these vulnerabilities exist. More importantly, we will explore practical strategies to identify, mitigate, and protect against these risks in business applications.
Audience Takeaways:
- Discover ways in which AI models are commonly hacked
- Understand the reasons why AI models are vulnerable to manipulation
- Learn practical strategies for mitigating adversarial AI risks within your organization
- Develop a deeper appreciation for human critical thinking as an essential complement to AI
This session will provide both technical and non-technical professionals with actionable insights to navigate the increasingly complex landscape of AI security and risk management.
This session will be recorded
