Armadin's $2.5bn price is for permission to attack; Quantum X sells a label
Armadin raised $255.5 million at a valuation above $2.5 billion seven months after launch, 5.6 times the $445 million investors have put in. The asset is trust: customers letting an AI swarm attack their live systems. A Tel Aviv company with a $65 million market value shows what a quantum label alone is worth.
The cleverest thing about Armadin is a signature. Few companies will let an outside swarm of AI agents chain together weaknesses in their production systems until the cloud account falls over, and fewer still will hand that permission to a start-up nobody has heard of. Armadin has the permission because Kevin Mandia runs it. A $255.5 million Series B co-led by Andreessen Horowitz and Accel, announced on Thursday, values the company above $2.5 billion seven months after it came out of stealth. Investors are paying for a licence, with the software as the delivery vehicle.
Armadin launched in March with nearly $190 million. With the new round, total funding is $445 million, so the Thursday valuation is about 5.6 times all the money raised, or something like $360 million of paper value for each month of existence. With no revenue disclosed, the price is a claim about a market nobody has measured yet.
Mandia's thesis is blunt. "Offense is uniquely advantaged right now," he says. "AI lets an attacker find and chain weaknesses faster than any human team can respond." The past fortnight supports him. Z.ai's GLM-5.3 flagged 4,249 potential flaws across 389 software projects, and Google released Gemini 4 Argon first to vetted cyber defenders without its cyber guardrails. The gap between a disclosed bug and a working exploit is compressing, and the annual or quarterly penetration test is a photograph of a river.
The investor list reads like a trust ledger. Google's venture arm GV and In-Q-Tel, the CIA-backed fund that finds technology for the intelligence community, are returning backers, joined this round by Bain Capital Ventures and Redpoint. Kleiner Perkins, Menlo, 8VC and Ballistic also come back. A fund that has already written a cheque once, and a government-linked fund at that, is vouching for the thing before the customer does.
The remedy Armadin sells is to attack continuously. It runs what it calls agentic attack campaigns in production for Fortune 500 companies and government customers, with a swarm of agents that string low-severity weaknesses into validated kill chains, from an unauthenticated remote-code-execution bug through lateral movement to full cloud compromise. Defence is trained against the best offence available, every day. A scanner reports a list of problems; a swarm proves which ones lead somewhere, and a security chief can show a board the route an attacker would have taken.
That product needs a customer to say yes to something frightening. A tool that deliberately attacks live infrastructure can knock over a payments system, and it will do so in front of a chief executive who has to explain it. Buyers grant that permission to people whose names they have seen on incident reports. Mandia founded Mandiant, which Google bought in 2022 for $5.4 billion. WSJ Pro reports that he says his name opened doors, in a funding surge for early-stage companies fighting AI-powered threats. Most of those founders are asking for the same permission with a lesser name.
The strongest argument against Armadin's valuation is that autonomous red-teaming is a feature. CrowdStrike, Palo Alto Networks and Google, which owns Mandiant, already sell endpoint agents, firewalls and incident response to the same customers, and a bundled attack-simulation module priced at nothing would crush a stand-alone vendor. The argument is half right, and the half that is right sets the exit price. Platforms can build the software in a quarter. They cannot build a reputation that makes a Fortune 500 general counsel sign off on live attacks, and the cheapest way to acquire one is to buy the people who have it. Google has done it once already, when Mandia's firm was the name chief executives called after a breach.
That is why we expect at least one of the large security platforms to buy an autonomous red-teaming start-up by October 2027. Armadin is the largest candidate, but any rival with a few marquee customers would do. If Armadin needs another round before an acquirer calls, its next step-up should be smaller than this one. Going from roughly $190 million raised to $445 million in seven months, at a valuation that already assumes a category, leaves less room to double again, and the competitors who will arrive with similar software are already being funded.
At the other end of the market sits a company with a different asset. Quantum X Labs, based in Tel Aviv, announced on Thursday that it had filed a US patent application, still pending, for a technique called LayerQake. The company's release describes quantum stratified inference, amplitude amplification and Bayesian analysis applied to rare anomalous events, with telemetry from physical plants and sensor networks as the worked example, and says it works without increasing the number of qubits.
The shares were up about 6% at $4.90 by late morning in New York, valuing the company at around $65 million. By then only 3,427 shares had changed hands, about $17,000 of stock. The parent also owns Gix Media, which monetises internet traffic, and Metagramm, which makes grammar software, and it has just named Yossi Cohen president of its scientific advisory board. A filing is a place to start. It is a long way from a product, never mind a customer who has granted permission to use one.
Cybersecurity money is plentiful this autumn, and it goes to whoever can show a buyer that the thing works. Armadin has done that inside Fortune 500 networks, and its price reflects the scarcity. Late-stage investors who paid up this week are betting on that exit, and the venture-backed rivals that arrive with similar software will compete for the same few buyers. By the first quarter of 2027, with a rival's round or an acquirer's cheque, we will know whether a platform agrees that the trust is worth buying.