Kevin Mandia, the founder of Mandiant, which sold to Google for $5.4 billion in 2022, has raised $255.5 million for his new security startup Armadin at a valuation of more than $2.5 billion, the company announced on Thursday, according to TechCrunch. The Series B round was led by Andreessen Horowitz and Accel. It comes just six months after a $190 million Series A in March, and Armadin has now raised more than $445 million in total.
The figure that matters most here is not the $255.5 million. It is the interval. A company that moved from a $190 million Series A to a $255.5 million Series B at more than a $2.5 billion valuation inside six months is being funded on the strength of a thesis, not a decade of operating history. All valuation and raise figures are as announced by the company and reported by TechCrunch; no independent valuation has been published.
The thesis itself is legible to anyone who follows the broader security debate: the same agentic AI that enterprises are adopting can be turned against them, so the way organizations test their own defenses has to change. Armadin's stated answer is to replace point-in-time penetration testing with always-on agentic swarms that chain vulnerabilities together to break in, so holes are found and sealed before attackers, or rogue agents, can use the same techniques.
What is Armadin actually selling?
Strip away the funding news and the product is a reimagining of an old service. Traditional penetration testing works the way a fire drill does: hired testers attempt to break in on a schedule, report the weaknesses they find, and leave. Armadin, as described in the company's announcement, runs continuous agentic swarms instead — software agents that probe, chain vulnerabilities, and hack in on an ongoing basis rather than once a year.
The mechanism matters. A chained attack path is usually more dangerous than any single vulnerability, because it is the sequence that defeats layered defenses. Continuous testing that simulates those chains is, in principle, a closer match to how a real adversary works. That is the company's framing; the TechCrunch report describes the intent, and it does not include independent performance data on how well the swarms perform.
Why the investor list is part of the story
The round's composition says something about who expects agentic security to matter. Alongside Andreessen Horowitz and Accel, the round drew Bain Capital Ventures, Redpoint, 8VC, Ballistic Ventures, Google Ventures, In-Q-Tel, Kleiner Perkins, and Menlo Ventures, per TechCrunch. In-Q-Tel is listed among the investors in the TechCrunch report, and its presence is the detail a defense reader will notice in a company whose core product is autonomous offensive simulation.
That does not mean Armadin has any government contract, and the reporting does not say it does. But the buyer map is visible in the investor list. Enterprises are the stated first market. The intelligence and defense community is watching the same agentic-attack problem from the other side, which is why coverage of cyber forces and election-guard duty, such as our earlier piece on NSA and military cyber forces put on election guard duty, keeps converging on the same question: who finds the holes first. Readers following this should also see Hegseth puts NSA and military cyber forces on election guard duty weeks before the midterms. Readers following this should also see Hegseth puts NSA and military cyber forces on election guard duty weeks before the midterms.
What the $255.5 million round does not establish
Restraint is due here, and it is the useful kind. The announcement establishes that sophisticated investors believe the market is large and the team credible. It does not establish that agentic swarms outperform human red teams, that enterprises will pay for continuous testing at scale, or that the product works as described in production environments. None of those claims has been independently tested in the available reporting.
There is also a structural question the funding round cannot answer. Continuous attack simulation generates findings continuously, and findings only improve security if someone fixes them. That is the same constraint that limits every testing regime, human or automated. Readers who want the general method behind judging such claims can look at how public threat assessments can and cannot establish their conclusions: a capability claim and a verified outcome are different things. This connects to our earlier piece, What public threat assessments can and cannot establish, per published methods. This connects to our earlier piece, What public threat assessments can and cannot establish, per published methods.
Why a security funding round belongs in defense coverage
The connection is direct rather than metaphorical. Agentic AI changes the economics of offense: if software agents can find and chain vulnerabilities without a human operator's time, the cost of running an attack campaign falls. Defenses built around the assumption that a skilled human attacker is expensive to deploy are exposed to that shift. Armadin's pitch — that defenders should run the same agentic techniques against themselves first — is one commercial answer to it.
The pattern rhymes with physical defense, too. Layered protection only works if the layers are tested against realistic attack sequences, a point we made in an earlier explainer on what layered defenses are actually built to signal. Continuous adversarial testing is the cyber equivalent of exercising the assumptions inside a defense plan before an adversary does it for you.
What to watch next
Three things would firm up or deflate the thesis. First, named enterprise customers and published deployment results, which the current reporting does not include. Second, evidence on whether continuous agentic testing finds chained attack paths that annual human engagements miss. Third, whether government buyers — a constituency In-Q-Tel's participation may hint at — adopt the model. The $445 million raised so far buys the company time to answer those questions. It does not answer them.
