Horizon3 AI Inc. has established the Adversarial Intelligence Frontier Lab to leverage data from its NodeZero penetration testing service. The lab will be directed by Z. Berkay Celik, an associate professor of computer science at Purdue University. The facility aims to accelerate the development of cyber world models that provide red and blue agents with shared memory.
NodeZero currently processes approximately 30 billion tokens of anonymized data daily, collected from live production networks. This data tracks how defensive tools respond to security commands. The lab's initial research agenda includes studying coordinated AI agent attacks, deception techniques, and developing interruption and rollback mechanisms for autonomous security agents.
The company previously tested control mechanisms for security agents using Anthropic's Claude Sonnet 4 and Claude Haiku 4.5 models. Across more than 420 deployments, the system restricted AI actions to an approved list of tools. Additionally, the lab will oversee the company's Cyber Resilience Centers of Excellence, which focus on critical infrastructure.
Horizon3 reported reaching $100 million in annual recurring revenue in October, marking a 120% year-over-year growth. The platform has conducted over 350,000 production-safe tests, with 80% of clients transitioning from annual to monthly testing cycles. NodeZero is currently utilized by more than 7,000 organizations, including federal agencies, hospitals, and banks.
Founded in 2019, Horizon3 secured $250 million in funding this August, achieving a valuation exceeding $2 billion. The new research lab represents a strategic expansion of the company's capabilities in autonomous security, building upon years of data collection and research into software vulnerabilities and autonomous system defenses.
