Technology

Corma Raises $125 Million at a $1 Billion Valuation for AI Security Models

Martin HollowayPublished 2m ago3 min readBased on 1 source
Reading level
Corma Raises $125 Million at a $1 Billion Valuation for AI Security Models
Photo by Tima Miroshnichenko on Pexels

AI cybersecurity startup Corma has raised a $125 million Series A round that values the company at $1 billion, according to TechCrunch. The round was led by NEA, and Corma's total funding now stands at $185 million.

The Series A came about three months after Corma introduced its product. The company emerged from stealth, meaning it operated without a public launch, in August 2026, so the gap between its public debut and a billion-dollar valuation is short by any measure. Corma's product consists of AI models and agents trained for cybersecurity defense, according to the same report. In this context, an agent is software that carries out a sequence of actions on its own, such as pulling logs, querying systems, and drafting a response, rather than answering a single prompt.

Sequoia Capital, Khosla Ventures, and Coatue are among Corma's backers.

The broader context here is investor appetite for security products built on models rather than models bolted onto existing tooling. Security vendors have spent the last two years adding large language model features to SIEM platforms (which collect and correlate security logs), SOAR tools (which automate response workflows), and endpoint platforms (which monitor laptops and servers). Corma's pitch, as described, is different in kind: models and agents trained specifically for defensive work. That distinction matters to buyers because the question shifts from whether a vendor has an AI feature to whether its models measurably outperform a general-purpose model with retrieval (pulling in relevant documents) and tooling layered on top.

A $1 billion valuation at Series A indicates how much capital is willing to back that thesis before a long production track record exists. For the company, the money buys runway (the time it can operate before needing more funding) and hiring in a market where experienced security researchers and ML engineers are scarce. For incumbents, a well-funded competitor can now pursue enterprise deployments directly.

In my view, the test for Corma and its peers will not be funding announcements but operational outcomes: mean time to detect (how long an intrusion goes unnoticed), false-positive rates in analyst queues, and whether security teams can hand off routine investigations without losing auditability. Those are measurable, and they will decide whether purpose-built defensive models become a category or a feature set absorbed by larger platforms.

The optimistic reading is also the most practical one. If trained defensive agents can take over high-volume, repetitive triage, skilled analysts could spend more of their time on adversary tradecraft and architecture work, which is where the scarcity of talent hurts most. That outcome is not guaranteed, and the same agents can be misconfigured, over-trusted, or targeted. Still, the combination of a fast product cycle and a large first institutional round puts Corma among the companies that will be tested on that question over the next several quarters.