Corma Hits a $1 Billion Valuation Three Months After Leaving Stealth to Train Models for Security Defenders

Corma, which trains its own foundation models for defensive cybersecurity and deploys them as an autonomous "AI workforce" inside security teams, raised a $125 million Series A at a $1 billion valuation on November 16, 2026. NEA led, with Sequoia Capital, Khosla Ventures and Coatue returning, bringing total funding to $185 million.

An instructor coaches a line of young guards raising round shields against padded orange balls fed from a practice machine in a sunlit sapphire hall.
An instructor coaches a line of young guards raising round shields against padded orange balls fed from a practice machine in a sunlit sapphire hall.

Corma’s pitch is that general-purpose AI became a good attacker by accident, and defenders need models built on purpose. On November 16, 2026, the company announced a $125 million Series A at a $1 billion valuation, led by NEA, with existing investors Sequoia Capital, Khosla Ventures and Coatue participating again. The round brings Corma’s total funding to $185 million.

The timing is part of the story. Corma emerged from stealth in August 2026 with $60 million, a seed round its announcement says Sequoia led. Three months later, it has a valuation in ten figures.

A model trained on the defender’s data

Corma describes itself as a frontier AI lab for defensive cybersecurity. It trains its own foundation models and deploys them as the Corma Workforce, autonomous agents that work alongside a company’s security staff using the security and IT tools already in place.

NEA’s investment post explains the thesis. Frontier models, open or closed, learn from code and human language. Since a major attack path is finding and exploiting software flaws, NEA argues, general models “became excellent attackers almost by accident.” Defensive work is different: spotting, investigating and containing live attacks means digesting logs, events and configuration state, data that general models rarely see during training.

NEA highlights what it calls adversarial training for cybersecurity, inspired by the methods behind superhuman game-playing systems. The model is trained with large-scale reinforcement learning in realistic enterprise environments, against outcomes that can be checked, so it “learns by doing the work” rather than from text about security. NEA says Corma also treats security data as its own modality instead of forcing it into text or code, trains the model to pick out faint signals in large volumes of noise, and keeps it consistent across the thousands of decisions a single investigation can involve. These are investor descriptions; Corma has not published a technical report or benchmark results.

How the workforce operates

According to NEA, the Corma Workforce deploys in minutes and onboards like a new employee. It communicates through Slack, Teams and email, carries out tasks end to end, reports findings, and asks for approval when needed. NEA lists threat hunting, incident response, detection engineering and digital forensics among the areas customers are expanding it into.

NEA says Corma already has multiple production deployments at Fortune 500 and other large companies in healthcare, financial services, energy, critical infrastructure and retail, and that the system has autonomously detected, investigated and contained active attacks in real environments. It also claims customers’ security teams complete 10 times more work at lower cost per task. Corma has not named customers, and none of these figures have been independently verified.

”Black October” and a compute bottleneck

In a post signed by CEO Alon Pluda and the team, Corma ties its growth to what it says the industry called “Black October,” a wave of critical vulnerabilities and breaches in widely used software last month. Google’s threat intelligence group, in a separate report published the same day, notes that critical PHP flaws disclosed that month were patched on October 21, 2026, after early patched code had already circulated.

Corma says some of its enterprise customers already had the workforce in place when those incidents hit. Its main constraint now is compute. “Candidly, one of the biggest challenges we are continuously facing is to secure enough compute capacity to keep up with that demand,” the company wrote, adding that it is still working through a waitlist.

The new money is earmarked for scaling the foundation models and expanding compute, along with selective hiring. Corma also says it will share its next generation of models “soon,” without giving a date.

An 8200 veteran, DeepMind researchers and an NEA lead

NEA says Pluda served in an elite intelligence sub-unit within Israel’s 8200, and that the team combines experienced hackers with researchers from DeepMind, Microsoft and other frontier labs. Aaron Jacobson led the round for NEA, and NEA’s post is co-signed by Thomas Joshi and Lila Tretikov.

NEA compares the opportunity to coding: it believes Corma can do for defensive security “what coding models did for developers.” Whether the models live up to that comparison will be easier to judge once Corma publishes its next-generation release and says more about how it is evaluated.

Topics
  • #Corma
  • #Funding
  • #Cybersecurity
  • #Foundation Models
  • #AI Agents
Raj M

Author

Raj M

Contributor

AI Systems Architect is a seasoned technology leader with over 15 years of experience in the IT industry working with Fortune 500 companies. With a solid foundation in multi-agent systems, open-source LLM infrastructure, and enterprise deployment, he excels at building scalable production-grade AI platforms.