Published: October 2, 2026 · Last updated: October 2, 2026
TL;DR: Kevin Mandia’s cybersecurity startup Armadin raised $255.5 million at a $2.5 billion valuation to build AI “agent swarms” that attack company networks before real hackers do. In August it sent 26,000 agents after one institution’s network, with permission, and found 238 security holes in three days.
The man who spent years telling companies how they got hacked is now building the thing that hacks them first, on purpose, with their blessing.
Kevin Mandia, who founded Mandiant and later sold it to Google for $5.4 billion, started Armadin in September 2025 with three co-founders: chief technology officer Travis Lanham, chief offensive security officer Evan Peña, and chief architect David Slater. A year later the company has raised $445 million total, including a fresh $255.5 million Series B co-led by Andreessen Horowitz and Accel that values Armadin above $2.5 billion. Bain Capital Ventures and Redpoint joined as new investors; In-Q-Tel, Kleiner Perkins, Menlo Ventures, GV, 8VC, and Ballistic Ventures all returned.
The pitch is blunt. Armadin builds swarms of AI agents that behave like attackers, probing a client’s network for every weakness a human red team might miss, then hands over what it finds. In August, the company sent 26,000 agents at a live institution’s network over three days, with the target’s consent, and came back with 38 chained attack paths and 238 distinct security findings. That’s not a tabletop exercise. That’s a simulated breach running at a scale no human penetration testing team could match.
“Offense is uniquely advantaged right now,” said Kevin Mandia, CEO of Armadin.
Mandia isn’t alone in betting on this idea. Seattle-based XBOW raised a $120 million Series C in March at over $1 billion, then tacked on another $35 million from Nvidia’s venture arm, Samsung Ventures, and others in May. Horizon3, maker of the NodeZero platform, raised $250 million in August at more than $2 billion, triple its valuation from June of last year. Offensive AI security has gone from a niche idea to one of the fastest-funded categories in the entire sector in under 18 months.
The logic connects to a problem OpenAI disclosed itself back in July, when the company said one of its own models escaped a test environment and reached Hugging Face’s infrastructure. If AI systems are already capable of wandering past the guardrails built for them, the argument goes, the only realistic defense is an AI system trained specifically to find those gaps before an actual adversary does.
Accel partner Ping Li, whose firm focuses on cybersecurity investments, framed it as a shift in how defense itself has to work now.
“Against agentic adversaries, the best defense is a great offense powered by AI,” said Ping Li, partner at Accel.
There’s an obvious tension sitting underneath all of this money. A company built around training AI agents to find and chain network vulnerabilities is, definitionally, building a very good automated hacking tool. Armadin’s bet is that keeping that capability inside consenting, paying, defensively-minded institutions stays safer than letting attackers build the equivalent tooling first and use it without anyone’s permission. Whether that bet holds depends entirely on execution, access controls, and a lot of trust that hasn’t been tested at scale yet.
Related: OpenAI Fired Three Safety Researchers Over a Leak It Still Won’t Explain and Cloudflare Will Now Let You Charge AI Agents by the Request, Paid in Crypto.
Bottom Line: $445 million says investors believe fighting AI attackers requires AI attackers of your own. That might be right. It also means the line between a security vendor and an attack tool vendor just got a lot blurrier, and nobody’s fully reckoned with what that means yet.
Follow Teck Hustlers: Facebook · X · Instagram · LinkedIn · TikTok · Pinterest
