August 20th, 2026 @5:30 PM – Kevin Latchford – Cracking Claws: A Study of Autonomous Agent Runtimes 

Register:
This is a Hybrid meeting. A dinner meal will be served (Meal menu &pizza). REGISTER HERE for IN PERSON & ONLINE via MS Teams. Location: Microsoft Reston 11955 Freedom Dr., Reston, VA 20190
Location: Floor 2, Rooms 2.2C, 2.2D. Free garage parking directly across from location. Metro Station Accessible. In-person registration closes on Tuesday, August 18, 2026. Registrations received on August 19 and August 20, 2026 will be restricted to online participation via Microsoft Teams. No in-person attendance options will be available after August 18, 2026.

Abstract:
Agentic AI systems — software that gives large language models the ability to take actions, call tools, and operate autonomously on behalf of users — are moving fast from prototype to production. And as with every previous generation of powerful technology, the security community is playing catch-up.

This talk uses a set of real-world agentic frameworks as a case study in how that catch-up happens. We start with OpenClaw, one of the earlier serious attempts to build a general-purpose LLM agent, and follow the thread forward through its successors: the ZeroClaw and NanoClaw derivatives, which evolved the original design under the pressure of real deployment, and Hermes Agents, a ground-up redesign that had the luxury of learning from what came before.

By putting these systems side by side, we can trace how the security has evolved with each iteration and improvement — from the specific languages these tools are based on, toolchain skills used, trust boundaries between the agent and the model, and the wider implications of what happens when an attacker figures out they don’t need to break your infrastructure if they can just tell your AI to do it for them.

Whether you’re building agentic systems, evaluating them, or trying to govern their use inside your organization, this session will give you a concrete framework for understanding where we’ve been and where we’re going in agentic AI security.

Speaker Bios:
Kevin Latchford is a cybersecurity and AI practitioner with many years’ experience in the field. Starting in telecommunications, Kevin built skill sets in both networking and Linux systems before taking on his current work in cybersecurity and artificial intelligence. He is notable for his AI security work with Llama 3, which he and fellow practitioner, Alexander “Kai” Smith demonstrated successful jailbreaking techniques for testing guardrails on contemporary LLMs. A report on which has gained him interviews on Cybersecurity Today, The Cyber Coffee Hour, and Data Driven. He is currently working with Spencer Trask & Co., heading up their operations and projects in secure agentic AI solutions development.