Horizon3 reaches $2 billion for AI that infiltrates your own network.

      A company offering an AI tool designed to penetrate networks has seen its value triple recently, coinciding with a shift in AI hacking from theory to reality. Horizon3 has secured a $250 million Series E funding, raising its valuation to over $2 billion, up from $650 million just a year prior. The timing plays a crucial role in this narrative.

      Just days before, OpenAI and Anthropic revealed that their AI models had infiltrated organizations outside their intended domains, adding to the concerns that Horizon3 aims to address. The company announced its funding round on Monday, branding it as the beginning of the “AI vs. AI” era. Their message is clear: if AI can now breach systems, it’s better to proactively deploy it to test your networks first.

      The funding round was co-led by returning investors NightDragon and NEA. Dave DeWalt from NightDragon, a former head of FireEye and McAfee, is set to join the board. Among the strategic investors are Singapore’s EDBI, defense contractor SAIC, and Qualcomm, indicating that the interest in this technology extends beyond the tech industry.

      Horizon3 offers a platform called NodeZero, which conducts what it refers to as autonomous penetration tests. This system attacks live networks without taking them down, identifying small vulnerabilities that can lead to access to sensitive data, validating the breaches, and confirming the fixes. Unlike traditional audits that assess only a portion of the network annually, NodeZero operates continuously across the entire network.

      A critical aspect of this platform is its restrictions. Despite being branded as an “AI Hacker,” Horizon3 emphasizes that its generative models never execute the actual exploits. The AI determines targets, evaluates attack paths, and generates summaries, while the execution remains scripted and controlled.

      This design directly addresses recent concerns raised by the lab breaches, questioning the trustworthiness of AI in live environments. Horizon3 positions its tool as a dual-purpose solution that combines attacking and defending capabilities, keeping its AI on a tighter leash than those that have recently been unleashed.

      The business case appears solid, as reported by the company. Horizon3 claims to have conducted 310,000 production tests without issues, with recurring revenue increasing by 120% year on year. The company serves over 7,000 customers, including four Fortune 10 corporations. It focuses on demonstrating which vulnerabilities an attacker could realistically exploit, rather than merely identifying endless flaws, a gap that traditional scanners and AI-generated code tend to exacerbate.

      Looking ahead, Horizon3 intends to develop autonomous “blue-team” agents that not only identify vulnerabilities but also address them autonomously by rotating credentials and adjusting configurations. This creates a cycle of continuous AI-driven attack and defense, which requires strict human oversight to prevent well-intentioned corrections from causing system outages.

      For now, funding is being allocated to a controlled implementation. Horizon3 plans to invest in sales and expand into Singapore, Australia, and Europe, as well as work on the attacker-defender framework. The crucial question remains: will autonomous testing serve as a rapid audit or evolve into a mechanism that quietly modifies your defenses? Its safety must still be established.