Horizon3 announced the launch of its Adversarial Intelligence Frontier Lab to develop closed-loop systems that use evidence from production-safe NodeZero attack graphs to find and close real attack paths at machine speed. While most AI security efforts train primarily on sensor telemetry or test against digital twins, Horizon3’s Adversarial Intelligence Frontier Lab starts from verified, end-to-end attack paths that have already succeeded against live systems. Its research will focus on keeping these attack-and-defense systems safe, observable, and controllable in production.
“We’ve spent years collecting the attack-and-defense data required to research these systems. The next step is to use that verified attack-path evidence to improve how offensive and defensive agents work together in environments that actually matter,” said Snehal Antani, CEO and co-founder of Horizon3.
Research Leadership
Dr. Z. Berkay Celik serves as the inaugural director of the Adversarial Intelligence Frontier Lab. He holds an appointment as an Associate Professor of Computer Science at Purdue University and also serves as co-director of the Purdue Security Laboratory. His research investigates the security of complex software and systems, with an emphasis on vulnerability discovery, attack analysis, and the design and evaluation of defenses for agentic and autonomous systems. He has led multiple collaborative research projects with academic, industry, and government partners.
The Lab builds on Horizon3’s published research into AI attacker behavior, cyber deception, and autonomous defense, including:
- Honeyquest for LLMs: Rethinking Cyber Deception for AI Attackers
- Closing the Sim-to-Real Gap: An Evaluation Framework for Autonomous Cyber Defense Configuration of Commercial EDR
- Stable Agentic Control: Tool-Mediated LLM Architecture for Autonomous Cyber Defense
The Adversarial Intelligence Frontier Lab will initially pursue four lines of research:
- AI-enabled attacks and adversarial behavior: Using frontier models to accelerate vulnerability discovery and studying emerging attacker tools, tactics, and coordinated agent swarms. Research will examine how these threats operate, scale, and respond to deception, and what defenders can do to detect and disrupt them.
- Adaptive defense and attack-and-defense learning loops: Evaluating how AI-enabled attacks bypass existing controls and how coordinated offensive and defensive agents can accelerate the hack, fix, verify cycle. Research will explore adaptive deception and ways to safely share findings across environments to improve defenses, limit attacker reach, and shorten exposure windows.
- Safe, controllable, and measurable autonomy: Developing architectures and rigorous evaluation methods for autonomous security agents grounded in production-derived evidence from NodeZero. Research will examine scope constraints, interruption and rollback mechanisms, and reproducible measures of agent effectiveness, defensive outcomes, and safety.
- Cyber resilience and operational readiness: Studying how AI changes the protection of critical infrastructure and the conduct of cyber and defense operations, including coordination across operational domains. Research will also examine the skills, education, and training needed to prepare the future cyber workforce.
The Adversarial Intelligence Frontier Lab will also accelerate the training of customer-specific cyber world models that underpin the shared context and long-term memory layer for red and blue agents, enabling more effective coordination and persistent reasoning across attack and defense cycles.
Unique Training Data from Real Production Systems
Horizon3’s base of more than 7,000 customers, ranging from large banks and federal agencies to advanced manufacturers and local hospitals, provides a foundation for the Lab’s research. These organizations generate diverse, labeled training data from real production systems through hundreds of thousands of production-safe NodeZero tests.
Horizon3 estimates it is now collecting 30 billion tokens of anonymized, aggregated, and cleanly labeled attack-defend training data per day. The data includes every attacker command executed, whether those commands succeeded or failed, how defensive tools reacted to those commands, and the context of how key services are configured behind the firewall. Unlike datasets derived primarily from sensor telemetry, synthetic environments, or digital twins, this produces a detailed record of real attack-defend interactions in live environments.
That combination of scale, diversity, volume, and ground-truth context gives the Lab a distinctive foundation for advancing frontier research in closed-loop offensive and defensive AI.
Critical Infrastructure Resilience
The Adversarial Intelligence Frontier Lab will oversee Horizon3’s Cyber Resilience Centers of Excellence focused on critical infrastructure. These centers will translate the Lab’s research into practical advances that strengthen the resilience of essential systems across key sectors, applying production-validated attack-path intelligence and customer-specific cyber world models to environments where downtime or compromise carries national or economic consequences.
