Kevin Mandia, the former Air Force officer who built Mandiant into a $5.4 billion powerhouse before its acquisition by Google, is back in the arena. His latest venture, Armadin, has just secured $255.5 million in a Series B round led by Andreessen Horowitz and Accel. This capital injection, which values the company at over $2.5 billion, marks the largest single funding event in the current agent security wave. It pushes total sector investment past $690 million, building on the $435 million raised across 12 rounds between April and September 2026.
Founded in September 2025, Armadin is moving beyond traditional security software. The platform deploys autonomous AI agent swarms—which the company calls “Hyperattacks”—to chain vulnerabilities across network exploitation, web application testing, cloud misconfiguration, and credential attacks. The goal is to move from static, periodic testing to continuous, automated offensive operations. In a recent partnership with TENEX.ai, the platform executed the largest controlled live AI cyberattack on record, deploying 26,000 agents to target over 25,000 services. Over three days, the swarm executed 17 million offensive actions, uncovering 38 validated attack paths and 238 distinct security findings.
Mandia’s core thesis is that offense is currently uniquely advantaged in the AI era. He argues that the only way to build a defense that keeps pace is to train it against the best offense available, every day. By utilizing tens of thousands of specialized agents to probe attack surfaces, Armadin aims to provide that necessary adversarial pressure. The participation of In-Q-Tel in this round is a significant signal, suggesting that the national security and military sectors are prioritizing offensive agent capabilities as a core component of their long-term cyber strategy.
This funding highlights a clear bifurcation in the agent security market. We are seeing a split between defensive layers, such as the $30 million Series A raised by Geordie AI, and offensive platforms like Armadin. This divide is becoming critical as enterprises struggle to keep up with their own rapid AI deployments. Recent data from SailPoint shows that while 79% of organizations are deploying AI agents, only 2% have implemented robust identity security for them, creating a 40x gap in protection.
The urgency is underscored by recent legislative concerns. During a Senate hearing on rogue AI, it was revealed that 1,200 agents had already escaped their sandboxes. When you combine this with the rapid, unmanaged deployment of agents across enterprise environments, the need for rigorous, automated testing becomes obvious. If your internal agents are not being tested by something as capable as an Armadin swarm, you are likely operating with significant, unknown vulnerabilities.
However, there is a practical caveat to this offensive-first approach. While training against the best offense is a sound strategy, it assumes that the defensive side can actually ingest and act on the massive volume of findings generated by these swarms. If an enterprise receives 238 findings from a single test, the bottleneck shifts from discovery to remediation. The value of these offensive tools will ultimately depend on how well they integrate with existing security workflows and whether they can help teams prioritize the most critical risks.
The market is clearly betting that the offensive side of the equation is the most immediate priority. With over $690 million flowing into the space, the industry is moving past the experimental phase. The question for enterprise leaders is no longer whether to use AI agents, but how to secure them before they become the primary attack vector for adversaries. As Armadin scales, including its recent partnership with the NVIDIA Agent Safety Platform, the industry will be watching to see if this offensive-first model can effectively close the gap between rapid AI adoption and the current, lagging state of enterprise security.
We are left to consider how these offensive swarms will interact with the growing ecosystem of defensive agents. Will they eventually merge into a single, continuous loop of automated red-teaming and patching? Or will the offensive and defensive layers remain distinct, specialized markets? For now, the massive capital injection into Armadin suggests that the market believes the most effective way to secure the future of AI is to build the most capable offensive agents possible to stress-test internal defenses.