Deception remains effective against AI-driven attackers.
Adversarial Intelligence
Frontier Lab
Studying how agents reason, act, and adapt across offensive and defensive operations.
Explore researchMission
Advance the science of adversarial intelligence for security
The Adversarial Intelligence Frontier Lab conducts rigorous research at the nexus of agentic systems and cybersecurity. We study the behavior, capabilities, and limitations of offensive and defensive agents; develop new methods for secure agentic systems; and establish frameworks to evaluate their reliability, security, and operational impact.
Our research is grounded in realistic security environments, attack paths, and defensive telemetry. We aim to translate scientific advances into measurable improvements in security systems and share findings, methods, and resources with the broader research and security communities.
Who we are
Research built on how attacks actually unfold.
Our research is built on Horizon3’s attack-fix-verify approach to security. This approach provides a strong foundation for studying attack paths, agent behavior, defensive controls, and the interaction between offensive and defensive systems.
We connect controlled experimentation with evidence from real-world security environments. This enables us to analyze how attacks progress, examine the decisions agents make, identify where defenses succeed or fail, and evaluate whether proposed changes measurably alter security outcomes.
Focus areas
Three questions we are working on now.
Three directions anchor the Lab today:
Cyber deception
Understanding how offensive agents respond to deception.
We study decoy infrastructure designed to detect, redirect, and contain offensive agents. Analyzing how these agents interact with deceptive systems reveals attack objectives, characterizes decision patterns, and provides signals that support attribution.
Governed Agentic Workflows
Ensuring safe & accountable agentic operation in security environments.
We study how multiple agents can collaborate to develop, test, and refine new security capabilities. To keep these multi-agent systems bounded and accountable, we design verifiable safety constraints and control architectures that restrict agent actions and prevent unbounded execution.
Precision defense
Translating offensive findings into validated defensive actions.
We analyze attack root causes to design targeted remediations across multiple infrastructure levels. By retesting the environment, we verify that the applied defense is correct and successfully mitigates the vulnerability or breaks the attack path.
Join the Lab
Help build the next generation of offensive and defensive security.
Horizon3 is expanding its research organization and is looking for researchers and engineers who want to work on autonomous attack and defense.
“We built NodeZero to show defenders how attackers can reach critical systems and whether a fix breaks that path. Now we’re researching how offensive and defensive agents can learn from those results, so teams can move faster from a proven attack to a verified fix. That is the next step in the hack, fix, verify loop.”
Snehal Antani
Co-founder and CEO, Horizon3