AI / Funding

🔥 Corma Hit $1 Billion in Three Months. Its Own War Games Explain Why.

Corma, a defensive-cybersecurity AI lab, raised a $125M Series A at a $1bn valuation about three months after leaving stealth.

Corma Hit $1 Billion in Three Months. Its Own War Games Explain Why. — Tech Arcade
Photo: Tyler / Unsplash

A cybersecurity company most people had never heard of in July is now worth a billion dollars. It took about three months.

Let’s get the number on the table first: $125 million. That’s the Series A that Corma, a defensive-cybersecurity AI lab split between Tel Aviv and San Francisco, closed this week at a $1 billion valuation (TechCrunch). The round was led by NEA, with Sequoia, Coatue and Khosla Ventures already on the cap table. Add it to the $60 million seed the company raised in August and Corma has pulled in $185 million total since it came out of stealth (Tech Monitor).

Three months from product launch to unicorn is fast even by 2026 standards. The interesting part isn’t the speed, though. It’s why investors paid up.

Corma ran its own war games, pitting AI against AI inside simulated Fortune 500 networks. The attackers won 88% of the time. The defenders caught 12%. That gap is the entire pitch.

🧠 Why This Matters

For two years the story about AI and security has been lopsided in one direction: the bad guys got better tools. Agentic models can now scan, probe, chain exploits and move laterally through a network at machine speed, and they don’t get tired at 3am. What has been missing is a defender built to the same spec.

Corma’s bet is that you can’t bolt AI onto a legacy security product and call it even. You need a foundation model trained from scratch for defence, running agents that work like extra staff on the security team rather than another dashboard nobody reads. Founder and CEO Alon Pluda, whom Sequoia describes as “one of the most elite hackers in the world,” put the stakes plainly:

“The race to general intelligence in cybersecurity has already begun, and the attackers have a significant head start.” — Alon Pluda, Corma CEO (Pulse 2.0)

If that head start is real, every enterprise with a network is a potential customer. That’s the math NEA is buying.

📊 Deep Dive

The reason Corma’s own simulations matter is that they tested the uncomfortable case: what happens when attacker and defender have the same AI? In red-team/blue-team runs where both sides used leading models from the likes of OpenAI and Anthropic, the attacker planted a backdoor and the defender, an identical copy of the same model, missed it 78% of the time (Sequoia). Sequoia’s one-line summary of the finding is the part that should worry CISOs:

“Holding model quality constant, attackers will always have an advantage.” — Sequoia Capital

Here’s how the round stacks up against the moment Corma stepped out of stealth:

  • Seed (Aug 2026): $60M, led by Sequoia, at launch out of stealth
  • Series A (Nov 2026): $125M, led by NEA, roughly 90 days later
  • Valuation: from private seed to $1 billion in a single quarter
  • Total raised: $185 million
  • Backers stacked: NEA, Sequoia, Coatue, Khosla Ventures
  • Footprint: two offices (Tel Aviv, San Francisco), a team drawn from Google DeepMind research and Israel’s Unit 8200

Corma also points to one customer case to show the model does something in the wild: its agent reportedly found and shut down an active attacker campaign inside a client’s network within its first hour, a campaign the in-house team had missed for 52 days (Sequoia). Worth noting that figure, like the 88% and the 78%, is company-reported, not independently audited.

⚠️ The Catch

That caveat is the catch, and it’s a big one. Nearly every eye-catching number here comes from Corma’s own lab or its own customers. The 88% attacker win rate, the 78% miss rate, the 52-day save, the claim that early clients cut response times and widened coverage by multiples: all of it is self-reported, and a company raising at a billion-dollar valuation has every incentive to frame the threat as severe and its own fix as working.

There’s also the crowd problem. Corma isn’t alone in spotting that AI defence is a gap. Kevin Mandia, the founder of Mandiant, is building a rival called Armadin, and Anthropic, OpenAI and Microsoft are all pushing security products of their own (TechCrunch). A billion-dollar valuation three months in prices Corma for a win it hasn’t secured yet. And the deeper tension is structural: if a smarter defender model becomes available, a smarter attacker model usually arrives alongside it. Corma is selling a race, not a finish line.

🎯 What Happens Next

The next 12 months are about proof. Corma now has the capital to scale its model and its agent fleet, and with Fortune 100 and Fortune 500 logos reportedly already in the mix, it has the design partners to generate real-world results. The question is whether independent data starts to back the lab-bench claims, because enterprise security buyers are a famously hard sell and “trust our internal simulations” only gets you through the first meeting.

Expect the competition to get louder and the valuations across AI-native security to keep climbing. When Sequoia, NEA, Coatue and Khosla are all circling the same thesis, the sector tends to get expensive before it gets proven.

🧩 Bigger Picture

Strip away the funding drama and Corma is a bet on a specific, slightly bleak idea: that cybersecurity is becoming a contest between two AIs, and whoever owns the better-trained one wins. That’s why Sequoia keeps stressing “sovereign AI” — owning the model weights so you aren’t bound by another lab’s rules on security work, and so inference stays cheap enough to run defence at scale.

If that framing holds, security stops being a feature you buy and becomes a model you deploy, priced and updated like any other frontier AI. If it doesn’t, Corma is a very expensive demo. Either way, a startup that didn’t exist publicly in July is now a billion-dollar referendum on whether machines can out-defend the machines attacking them.

The attackers have an 88% head start. Corma just raised $125 million to argue that number can be flipped.


Sources

❓ Quick answers

How much did Corma raise and at what valuation?

Corma raised a $125 million Series A at a $1 billion valuation, led by NEA, announced on 16 November 2026 (TechCrunch).

What does Corma do?

It builds a foundation model and AI agents purpose-trained for defensive cybersecurity, deployed like extra security staff. Founder and CEO Alon Pluda leads a team drawn from Google DeepMind and Israel's Unit 8200.

Why is this valuation controversial?

The $1bn price came roughly three months after launch, and headline figures like the 88% attacker win rate are from Corma's own simulations rather than independent audits, with rivals such as Mandiant founder Kevin Mandia's Armadin also entering the space.