AI-generated illustration of a shield beside a server rack and gold coins. It is not a picture of anything in this story. Illustration: Grok.
By news editor
Corma, a startup that trains AI foundation models for defensive cybersecurity, has raised a $125 million Series A at a $1 billion valuation, three months after it emerged from stealth with $60 million in funding. NEA led the round, with Sequoia Capital, Khosla Ventures and Coatue participating again, the company says in a blog post signed by chief executive Alon Pluda and the Corma team.
Total funding now stands at $185 million, according to the post. It does not say whether the $1 billion valuation is counted before or after the new money, and it gives no revenue figure.
Computing power is the first use the company names for the cash. The post states, "The new funding will allow us to accelerate the scaling of our foundation models and spike our compute capacity." Hiring comes second, and the company says it will expand its team selectively.
Capacity is also what Corma describes as the limit on its growth. The post states, "Candidly, one of the biggest challenges we are continuously facing is to secure enough compute capacity to keep up with that demand." It addresses prospective customers on a waitlist and says the company is working to expand access.
Foundation models are the large, general-purpose AI models that other software is built on. Corma's are trained for defense and sold as what the post calls the Corma Workforce, which it says works alongside security teams using their existing security tools. The company was founded in 2025 and is headquartered in Tel Aviv and San Francisco, SecurityWeek reports.
SiliconANGLE reports that the models find breaches by analyzing telemetry from firewalls, network monitoring software and other security applications, then generate remediation guidance, and that administrators can steer them through a Slack chatbot.
No customer is named. In its post, Corma says "some of the world’s leading enterprises" were already running the product in October, but it does not identify them. Neither the post nor the three news reports gives a price.
Performance claims come from the company as well. According to SiliconANGLE, Corma says its models can speed up some breach response workflows by a factor of 100, and that one of them outperformed several rival large language models on an evaluation called DefenseBench V5. SiliconANGLE describes that benchmark run as Corma's own, and no independent test is cited.
Demand, by the company's account, is tied to October. The post describes a wave of critical vulnerabilities and breaches in widely used software during October, which it says the industry called Black October. TechCrunch reports that interest in the product rose after October 21, when a critical flaw was found in the PHP programming language, another was found in Node.js, and LiteLLM disclosed a security incident.
Rivals are numerous. TechCrunch lists Kevin Mandia's startup Armadin along with Anthropic, OpenAI and Microsoft among the companies using AI for defense, and notes there are many other startups in the field. Corma's answer, in TechCrunch's telling, is that it is a frontier AI research lab focused on protection.
For buyers, the round does not yet change what can be purchased. Access remains limited by capacity, by the company's own account, and terms are not public.
The post says the next generation of Corma's foundation models will be shared soon, without giving a date. The details to watch for in that release are named customers, a published price and a benchmark run by someone other than the company.

Leave a Reply