FRIDAY, NOVEMBER 20, 2026 Tech news, minus the noise.
N News Cast
LATEST
Startups

Corma Raised $125 Million to Build AI That Only Plays Defense

Corma, a startup training foundation models purely for defensive cybersecurity, raised $125 million at a $1 billion valuation just three months after leaving stealth. Investors are betting that defense, not offense, is where AI security money gets made.

Corma Raised $125 Million to Build AI That Only Plays Defense
Source: Azaleos via Wikimedia Commons (CC BY-SA 3.0)

Corma raised $125 million in a Series A round at a $1 billion valuation, and it did it just three months after leaving stealth. The Corma Series A was led by NEA, with existing investors Sequoia Capital, Khosla Ventures and Coatue all putting in more money. Total funding now stands at $185 million, including a $60 million seed round in August led by Sequoia. For a company founded in 2025 with about 20 to 25 employees, that is a remarkable amount of belief in a short amount of time.

The bet is specific. Corma trains foundation models purely for defensive cybersecurity, then deploys them as an AI workforce that works inside a customer's existing security tools. The company says general-purpose AI models are better at attacking than defending, so it builds its own models from scratch, trained only on defense. The agents investigate alerts, correlate signals across tools, and take action alongside human security teams.

What the Corma Series A money buys

Three things, according to the company: bigger models, more computing capacity, and a limited amount of hiring. Computing is the binding constraint. Corma says demand from large enterprises already outstrips the model capacity it can serve, which is an unusual problem for a startup that is barely a year old and a useful signal about how hot this market is.

The pitch to customers is speed. Corma claims its models can speed up some breach response workflows by a factor of 100 while cutting costs, and it points to its own DefenseBench V5 evaluation to back that up. Those are company numbers, not independent results, and they should be read that way. But the customer list is doing real work for the story: Corma says its clients include Fortune 100 and Fortune 500 companies across healthcare, financial services, energy, retail and telecommunications, and its annual recurring revenue is estimated in the tens of millions of dollars.

The founder story helps explain the speed. CEO Alon Pluda previously co-founded Miggo Security and is a veteran of Israel's Unit 8200 signals intelligence unit. The team mixes frontier AI researchers, including people from Google DeepMind, with offensive and defensive security specialists. Corma runs from Tel Aviv and San Francisco, and that split is deliberate: the research talent and the enterprise customers live on different continents.

Why defense is suddenly fundable

Cybersecurity has a long-standing problem: the attackers adopted AI first. Phishing, vulnerability discovery and social engineering all got faster and cheaper with AI, while defense stayed mostly rules and dashboards. The October breach wave showed what that gap costs in practice, and the funding market has noticed. Startup funding has been flowing toward AI infrastructure all year, and security is now part of that flow.

Corma is not alone in this bet. The whole defensive AI category is filling up fast, from established vendors adding agents to startups built around them. What sets Corma apart is the foundation model angle: instead of wrapping someone else's model in security tooling, it trains its own model for defense from the ground up. That is expensive and slow, which is exactly why it needs $125 million, and it is also the moat if it works. A model trained only on defense should, in theory, be better at defense than a general model asked to moonlight as a security analyst.

The honest caveats

A $1 billion valuation for a company founded last year is a bet, not a measurement. Corma has not published revenue, customer counts or losses, and the 100x workflow claim comes from its own testing. The Ramp raise and the Modal raise show the same pattern across AI infrastructure: huge valuations on usage figures and customer logos, with profitability somewhere in the future.

There is also a real question about how much autonomy enterprises will actually hand to AI agents in security. Investigating an alert is one thing. Taking action, isolating machines, blocking accounts, is another, and most security teams will want a human in the loop for a long time. Corma's answer is that its agents work alongside humans and take direction through chat tools like Slack or Teams, which is the right framing, but the trust has to be earned one incident at a time.

Still, the direction of travel is hard to argue with. Attackers are automating, so defenders have to automate too, and the companies that train models specifically for defense have a credible claim to do it best. Corma went from stealth to unicorn in three months. The next three months will show whether the product can keep up with the valuation.

AS
News Cast Editorial Team

We cover AI, gadgets, startups and security — original reporting and explainers, written by humans, for humans. Got a tip? Reach us via the contact page.