Threat Hunting Summit Keynote: Defending AI: Organized Musings on Securing AI Agents for Cybersecurity

With Jason Haddix

Are you securing your AI agents, or just hoping the system prompt holds?

As AI agents move from prototypes into production, most security programs are still treating them like traditional apps or opaque black boxes – and that gap is where real risk is emerging.

Join Jason Haddix, CEO and “Hacker in Charge” at Arcanum Information Security, as he lays out a practical architecture for defending AI agents based on real-world assessments and organizing controls across the full ecosystem: application design and tool boundaries, system prompt hardening, and intermediate guardrails and classifier-based protections.

He’ll cover how attacks actually manifest across these layers, the best security practices and common missteps seen in development teams, and where open-source defensive tooling fits today.

The focus of this talk is pragmatic: giving defenders a usable mental model, concrete configurations, and the tradeoffs they need to understand to make AI systems survivable in production.

If you’re building or defending AI systems, this is the playbook you’ll want to have.

Chat with your fellow Threat Hunting Summit attendees in the Antisyphon Discord server: https://discord.gg/antisyphon

---

 This talk is part of the Antisyphon Training Threat Hunting Summit, a free, six-hour, live virtual event designed to give you a practical, real-world look at how cyber threat hunters detect stealthy adversaries, investigate suspicious behavior, and turn discoveries into stronger defenses.

For those who want to go further, multiple hands-on, high-quality, and affordable training courses are available June 18-26 to help you sharpen your skills and become a more proactive, effective defender.

See the entire Antisyphon Training Course Catalog for affordable cybersecurity training! 

---