Trustworthy Agentic AI: Engineering Security, Compliance, and Governance from Ground Up
[2 Hours] Agentic AI systems are rapidly moving from experimentation into production — autonomously browsing the web, executing code, managing files, calling APIs, and collaborating with other agents. But as these systems grow more capable, their autonomy and stochastic behavior introduce a new class of risks that traditional security frameworks are ill-equipped to handle. Unlike conventional software, agentic AI makes dynamic, context-dependent decisions that are difficult to predict, audit, or constrain — creating vulnerabilities at every layer of the stack, from the LLM itself up through memory management, tool access, and the user interface.
This workshop explores why security, compliance, and governance represent the Achilles' heel of agentic AI. We examine the root causes — agent autonomy and the probabilistic nature of LLM reasoning — and trace how they manifest as concrete threats: prompt injection attacks, identity and API abuse, data leakage through persistent memory, cross-jurisdictional compliance violations, alignment drift, and the erosion of human oversight. We also look at how the absence of robust governance frameworks leaves organizations exposed as agents operate across tools, data sources, and organizational boundaries.
Founder & CEO - Ejento AI
Building the next generation of knowledge workers - powered by AI. Experienced AI professional, educator, and entrepreneur. Founder and lead instructor at Data Science Dojo. 10,000 aspiring data science professionals from 3000+ companies trained globally, and growing.
Workshop Overview
The session combines conceptual framing with a practical, hands-on component using Ejento AI, a security, compliance, and governance-first agentic platform. Attendees will leave with both a structured understanding of the problem space and direct experience applying mitigations in a real agentic environment.
Agenda
-
Getting Started: Introduction to the session and speaker background. We set up the central thesis: that the same properties making agentic AI powerful — autonomy, broad tool access, persistent memory, and multi-agent collaboration — are precisely what make it difficult to secure and govern.
-
Agentic AI Architecture: A layered walkthrough of how agentic systems are built, from the LLM core through prompting and routing, context and memory management, cognition and planning, infrastructure, and the user-facing layer. We examine how each layer introduces its own distinct attack surface and compliance exposure.
-
Risk with Autonomy: A focused look at the two underlying drivers of agentic AI risk: autonomy (agents taking actions, accessing data, and delegating to other agents with limited human supervision) and stochastic nature (LLM reasoning and decision-making is probabilistic, making deterministic security guarantees fundamentally difficult to achieve).
-
Security, Compliance & Governance Challenges: A structured deep dive into the three domains. Security covers prompt injection, context poisoning, API and identity abuse, and data leakage via tools and memory. Compliance covers data sovereignty across jurisdictions (GDPR, HIPAA, CCPA), auditability of dynamic agent decisions, and model and data lineage tracking. Governance covers policy enforcement across heterogeneous agent networks, alignment drift over time, and calibrating when to escalate decisions to a human.
-
Hands-On Case Study with Ejento AI: Attendees build a working agent on the Ejento AI platform, then iteratively harden it. Starting from a baseline agent with no controls, participants add input/output guardrails, role-based access control, memory redaction and scoped persistence, compliance validators and explainability logging. Each step demonstrates how the absence of a control creates a tangible vulnerability — and how adding it closes the gap.
Time and Location
March 31, 2026
3:15pm - 5:15pm
Cobb Galleria
Workshop Requirements
-
AI practitioners and researchers.
-
Developers seeking to transition into advanced agent-building roles.
-
Organizations looking to implement custom AI solutions.