Stop by the AFCEA table to pick up your name badge.

Alan Johnson is the VP of Strategic Development for GA-i3

Don’t Stop Me Now: Running Fast with Scissors in the Age of AI

AI is accelerating how we build software, analyze data, and deliver value—and challenging how organizations operate. At GA-I3, we call this “running fast with scissors”: embracing powerful tools with the judgment to use them effectively.

This talk explores the technical, organizational, operational, and cultural implications of AI adoption, with particular attention to companies delivering software and data science services to government. As government pursues more commercial solutions and agile acquisition approaches, providers must rethink delivery, pricing, workforce development, and the path from prototype to mission use.

Technical and nontechnical professionals will leave with practical ideas for turning AI-driven speed into dependable capability and measurable outcomes.

Hunter Moore is a Founder and the CTO for Hardshell

Measuring Data Exposure in Retrieval-Augmented AI Systems

Organizations across defense and government are connecting language models to internal document repositories. The security review for those systems typically covers the model and stops short of the retrieval layer and the knowledge base behind it, which is where the sensitive data actually sits.

This talk presents measured results from running published extraction attacks against a controlled RAG knowledge base seeded with tracked canary documents, then re-running those attacks against five published defenses to quantify what each one provides. Several widely deployed mitigations produce almost no measurable reduction in exposure. The defenses that do work impose costs on answer quality, and that tradeoff is the decision most programs have not yet made deliberately.

Attendees will leave with a method for assessing exposure in their own deployments, a comparison of current defenses measured against a common baseline, and specific questions to put to vendors claiming their products prevent data exfiltration.

Cory Everington, VP of Operations for DroneTrace

A Downed Drone Still Has a Story…

Most of that story is thrown away with the wreckage.

C-UAS investment has concentrated on one problem: defeating the system in the air.

Detect, track, identify, defeat.

That focus has produced real capability, particularly against one-way attack UxS in the CENTCOM and EUCOM AORs. But defeat ends the engagement. It doesn't end the mission.

Every downed or captured UxS carries data: flight logs, component sourcing, communications artifacts, media payloads. Left unexploited, that data disappears with the wreckage. Collected and analyzed, it maps the network behind the system — the supply chain, the operators, the tactics repeating across incidents.

This talk argues for extending the C-UAS cycle past defeat to include three additional steps: extract, analyze, disseminate. Drawing on patterns observed across recent CENTCOM and EUCOM engagements, I'll walk through what that extension looks like in practice: what data a downed system actually yields, how flight-pattern and supply-chain analysis turns individual incidents into network attribution, and where that intelligence needs to move to be useful at the tactical edge and beyond it.

Take time to network with speakers and attendees, and check out the booths at the event.

Steven Yates is the Founder and CEO of Federant

Governing Autonomous AI at the DDIL Edge: Who's Solving It, and Where the Data Lives

Edge and physical AI quietly assume that "mostly connected" means "mostly the same as always connected."

It does not.

When autonomous machines make consequential decisions in DDIL environments (denied, degraded, intermittent, and limited), the most important data is secure evidence of what the AI actually did while disconnected: what ran, what changed, who authorized it, and whether its actions were within operator-authorized bounds. In most deployments today, that data either does not exist, is not collected when the link is down, or sits in proprietary infrastructure the operator does not control.

This talk surveys the standards and open source efforts converging on the problem, including the new AI Deployment Software workstream at the Open Compute Project (which the speaker co-leads), adjacent projects across the Linux Foundation and IETF, and emerging defense and civilian regulations.

These efforts converge on the same missing layer, which is common infrastructure around the model that enforces operator policy at runtime, generates verifiable evidence of what the AI did, and makes autonomous AI deployable, observable, resilient, manageable, and secure. The goals are interoperability and true mission autonomy.

The speaker's Charlottesville-based company, Federant, is building this layer and testing it in live civilian deployments now.

Michael Prichard is the Founder and CEO of Shift AI

The Warrior Developer: Empowering Mission-Driven Software Creation at the Edge of Need

The next asymmetric advantage in defense and intelligence won't come from a program of record — it will come from the analyst, the operator, or the logistics specialist who builds their own tool because they understand the problem better than anyone in a faraway acquisition office.

AI coding agents have fundamentally changed who can build software. Today, a non-engineer with domain expertise and the right platform can create production-grade internal applications in hours. The challenge for defense and intelligence organizations isn't capability — it's control. How do you empower your people to build without exposing classified data, violating security policy, or creating ungoverned shadow IT that becomes a liability?

This talk introduces the concept of the Warrior Developer — the mission-focused practitioner who can now build, test, and deploy operational software without a single line of hand-written code. Drawing on the commercial emergence of the "Citizen Developer" movement and the governance frameworks emerging to support it, we explore how defense organizations can apply these same patterns in a security-first context: connecting AI coding agents only to approved data sources, enforcing classification and policy boundaries at the platform level, and producing fully auditable, IT-sanctioned applications — all without requiring engineering expertise from the warfighter.

Attendees will leave with a framework for thinking about governed AI-assisted development inside classified and sensitive environments, and a view into the tooling being built today to make the Warrior Developer a safe and scalable reality.

Moderator: Andrew Schoka, CEO, Hardshell

Panelists:
- Hunter Moore, Hardshell
- Steve Yates, Federant
- Cory Everington, DroneTrace
- Michael Prichard, Shift AI