Staging environment

You Can’t Patch a Brain: Basic Challenges of AI Security

Hosted by Mahesh Yadav and Dominique Wimmer

Fri, Feb 27, 2026

5:00 PM UTC (45 minutes)

Virtual (Zoom)

Free to join

Invite your network

Go deeper with a course

Agentic AI Product Management Certification
Mahesh Yadav
View syllabus

What you'll learn

The Illusion of Guardrails

Discover why current industry security filters are often "snake oil" that fail against adaptive attacks

Why You Can't Patch a Brain

Explore the "original sin" of LLM architecture—the concatenation of trusted and untrusted input

Engineering Robustness with CaMeL - beyond "scaffolding"

Principled defensive architecture that separates a system's control flow from untrusted data

Why this topic matters

As AI agents gain autonomy, weak guardrails turn small prompt hacks into serious security and safety risks. Understanding why traditional defenses fail—and how to design systems that separate control from untrusted data—helps you build safer AI by design. This lesson grows you from tool-user to systems thinker, capable of evaluating real-world risk in agentic AI products.

You'll learn from

Mahesh Yadav

Ex-GenAI Product Lead at MAANG Firms l AI PM Coach l 10k+ Alumni

Mahesh has 20 years of experience in building products at Google, Meta, Microsoft, and AWS AI teams. Mahesh has worked in all layers of the AI stack, from AI chips to LLM and has a deep understanding of how using AI agents companies ship value to customers. His work on AI has been featured at the Nvidia GTC conference, Microsoft Build, and Meta blogs.:

His mentorship has helped various students build real-time products & careers in the Agentic AI PM space.

Whether you're a hobbyist or a professional looking to get a grasp on GenAI Product Management, feel free to join our channels for more such sessions


Dominique Wimmer

Product @NYC OTI, Ex-Meta/Google, AI | Safety | Security

Dominique Wimmer has more than a decade of experience building products at Meta and Google, and most recently at the NYC Office of Technology & Innovation. As a Lead Product Manager she builds critical infrastructure for all New Yorkers’ service needs. In her career, she has built AI-powered trust & safety systems at scale, including reducing harmful content through ML-based detection and red-teaming GenAI products. Additionally, currently pursuing a Master's in Cybersecurity at NYU (part-time), Dominique mentors AI product managers on building safe, agentic AI systems for the Agentic AI Institute and serves as a Cohort Leader for Stanford's Ethics, Technology & Public Policy program. Her work bridges the gap between responsible AI principles and real-world implementation at billion-user scale.

Previously at

Google
Meta
Amazon Web Services
Microsoft

Sign up to join this lesson

By continuing, you agree to Maven's Terms and Privacy Policy.