Corma Emerges From Stealth With $60 Million From Sequoia to Train AI Models Built for Defense, Not Attack

Corma, a cybersecurity AI startup operating out of Tel Aviv and San Francisco, has come out of stealth with a $60 million seed round led by Sequoia Capital, with Khosla Ventures and Coatue joining, Fortune reported. The round is unusually large for a seed stage company and values the year-old startup at an undisclosed figure.

The company's thesis starts from an uncomfortable observation about the current generation of frontier models. CEO Alon Pluda told Fortune that the most capable models from OpenAI, Anthropic and Google are, in effect, trained to be good at offense: writing and optimizing code, identifying vulnerabilities, and reasoning step by step through a system - exactly the skills that translate directly to attacking infrastructure. Corma trains its models on the traits defense actually needs, which Pluda says have little to do with writing code: parsing logs, running audits, and finding the one anomalous event in a haystack, reliably, across thousands of operations.

The startup says it deployed its first model roughly six weeks ago to organizations across healthcare, financial services, energy, critical infrastructure and retail, including numerous Fortune 100 and Fortune 500 companies. It claims threat response time at adopting organizations has been cut by 94 percent - a company-reported figure that has not been independently verified. Sequoia partner Shaun Maguire said in the announcement that "agentic AI has given attackers a structural speed advantage, and Alon's hacker talents combined with Corma's frontier AI research can help enterprises counter these threats at scale."

The framing echoes what security teams have been living since AI-accelerated attacks started landing at machine speed: the incident that unfolded when an AI model was used to breach Hugging Face moved faster than any human-driven process could respond. Defense has been the afterthought in model training; Corma's bet is that the layer of intelligence built specifically for defenders becomes a necessary purchase for every large organization, the way endpoint detection did a decade ago.

The name, Pluda explained to Fortune, comes from Tolkien: "corma" is Elvish for "ring" - the all-powerful artifact in The Lord of the Rings - and Corma's product is meant to be that ring, "but this time for the defense." Whether a single seed-stage company can outbuild the security teams of the frontier labs themselves is an open question, but the round's size suggests Sequoia and its co-investors believe the defensive layer of AI security is about to become its own category rather than a feature.