Armadin Series B: $255.5M Bet on AI Agent Swarm Security Testing

A Big Round, Fast

Armadin, the AI security startup co-founded by Mandiant founder Kevin Mandia, has raised a $255.5 million Series B at a valuation above $2.5 billion.5 Andreessen Horowitz and Accel co-led the round. Bain Capital Ventures, Redpoint, 8VC, Ballistic Ventures, Google Ventures, In-Q-Tel, Kleiner Perkins and Menlo Ventures also took part.5 The financing came six months after a $190 million Series A in March. Total funding now stands at more than $445 million.512

The company was founded in 2025 in Palo Alto by Mandia, David Slater and Evan Pena.3 It has reached a multibillion-dollar valuation in roughly a year. Mandia's track record explains much of that speed. He sold Mandiant to Google for $5.4 billion in 2022.5 That history gives investors a reason to back him early, before the product has been tested at scale in public.

What Armadin Actually Does

Armadin's pitch is to replace periodic penetration testing with continuous automated attack testing. In a traditional pen test, a hired team tries to break in over a fixed window and then writes up what it found. Armadin instead runs swarms of AI agents that keep probing an organization's systems. The agents chain individual vulnerabilities together into working attack paths.51 The output is evidence that defenders can use to fix problems.4 The goal is to close gaps before attackers find them, including attackers who use agentic AI themselves.5

The company describes this as a change in the economics of security. Vulnerability scanners can list weaknesses across a whole environment, while pen tests offer depth only at a single point in time.2 Armadin wants to offer depth on a continuous basis. Mandia has also argued that offensive AI will shrink attack execution times from days to minutes.3 If that is true, an annual or quarterly test leaves a long gap between checks.

The 26,000-Agent Claim

The most striking detail is a disclosed August exercise. According to the company, it deployed 26,000 agents across more than 25,000 services and found 38 validated attack paths.1 One outlet reporting the figures adds two caveats. The numbers are reported by the company and vendor, and the client institution is not named.1 Armadin also says it already runs campaigns for Fortune 500 and government customers.21

The exercise does show what a swarm architecture is meant to do: attack in parallel across a very large environment. It does not show how well the swarm performs. Several key questions go unanswered:

• Without a named customer, the results cannot be checked independently.
• There is no baseline, so it is unclear how many of the 38 paths a human red team or an existing tool would also have found.
• No false-positive rate has been disclosed.

The figures read as marketing evidence for scale, not as proof of effectiveness.

A Crowded Field and a Trust Problem

The coverage agrees on the funding facts but differs in tone. TechCrunch and Konsulteer largely present the round and the company's product description.52 Remio stresses that Armadin is entering an established market. Horizon3.ai, Pentera, XBOW and other security validation vendors already automate parts of offensive testing.4 Tracxn's competitor data shows XBOW, an AI-driven autonomous pen-testing company, at Series C with about $270 million raised. Armadin has already passed that figure in total funding at an earlier stage.3

Remio also frames the central challenge as trust, not technical ability. A swarm of autonomous attackers running continuously inside customer environments must not disrupt production systems. It must not flood security teams with alerts they cannot work through. And it must not add new risk of its own.4 An always-on offensive system needs broad access, so the controls around it have to be at least as strong as the defenses it is testing.

Reading the Moment

The size and speed of the round, two large raises in six months, suggest investors are betting on Mandia's reputation and on the agent-security theme as much as on proven results. In-Q-Tel's participation hints at interest from the government and intelligence community.5 Its presence is a signal of interest, not evidence that the product works.

Armadin's thesis is reasonable. If attackers adopt AI agents, testing on a fixed calendar will likely fall behind. But the company's best-known result so far is a self-reported exercise with an unnamed client. The next proof points that matter would be:

• named customers,
• third-party evaluations,
• head-to-head comparisons against Horizon3.ai, Pentera and XBOW, and
• transparency about false positives and operational safety.

Until those appear, the $2.5 billion valuation is a bet on the founder and the theme more than a reflection of demonstrated results.