wasl.tech

Corma AI Cybersecurity Startup Reaches $1 Billion Valuation in Three Months

Three months. That is all it took for Corma, an AI cybersecurity startup most people had never heard of in August, to become a billion-dollar company. On November 16, the Tel Aviv and San Francisco startup announced a $125 million Series A round at a $1 billion valuation, led by NEA.

For a company with roughly 25 employees, the numbers are startling. Corma has now raised $185 million in total funding since August, when it emerged from stealth with a $60 million seed round led by Sequoia Capital.

The speed of the rise says as much about the moment as it does about the company. Cybersecurity teams are drowning in alerts, attackers are using AI to move faster, and investors are racing to back whoever promises to defend with the same technology.

Corma’s pitch is simple: build AI models trained specifically for defense, then let them do the security work that humans no longer have time for.

From Stealth to Unicorn in Three Months

Corma was founded in 2025 by Alon Pluda, a graduate of Israel’s elite Psagot program and an alumnus of Unit 8200, the military intelligence unit that has produced a generation of cybersecurity founders.

The company spent its first year quietly building. Then, on August 10, 2026, it stepped out of stealth with $60 million in seed funding and a bold claim: it was training foundation models for defensive cybersecurity, the way frontier labs train models for coding or conversation.

The seed round alone would have been a strong debut for a young startup. What happened next was rarer. Within three months, demand from large organizations was so strong that the company says securing enough computing power became one of its main challenges.

NEA led the new round, with existing investors Sequoia Capital, Khosla Ventures and Coatue all returning. The message from the cap table is clear: the investors who saw Corma first wanted more of it.

According to TechCrunch, the round closed just three months after Corma introduced its product. In a funding market that has grown cautious about AI hype, a billion-dollar valuation this fast is a signal that something about the pitch is landing.

What the Corma Workforce Actually Does

At the heart of Corma’s offering is the Corma Workforce, a system of AI agents designed to work alongside human security teams rather than replace them outright.

The agents plug into the tools a company already uses: identity systems, cloud platforms, endpoint protection, SIEM software, and firewall and intrusion-detection systems. They investigate alerts, connect evidence across those systems, and take action.

Employees can assign tasks through Slack or Microsoft Teams, ask the system questions, review its findings, and approve its actions. The design keeps a human in the loop while the AI does the heavy lifting.

That matters because modern security work is mostly triage. Analysts spend their days sorting through thousands of alerts, most of them false alarms, looking for the handful that signal a real attack.

Corma argues that general-purpose AI models are not built for this job. In its launch statement, the company said improvements in coding and software reasoning map more directly to offensive security than to defense.

Attacking a system is a bounded problem: find one weakness and exploit it. Defending one means watching audit logs, events, configurations and network flows continuously, connecting faint signals across systems and long stretches of time.

That distinction is the foundation of Corma’s strategy. Instead of fine-tuning someone else’s model, it is training its own foundation models from scratch for the defensive workload.

Why Investors Are Betting on Defensive AI

The timing helps explain the valuation. The past two months have delivered a steady drumbeat of AI security failures that make Corma’s pitch easy to understand.

In October, Fireworks AI disclosed a breach that exposed customer datasets and model weights. Days later, the Railway platform reported stolen AI keys that racked up hundreds of thousands of dollars in fraudulent charges. Microsoft researchers then showed how a handful of reprogrammed neurons could turn an AI agent rogue.

Each incident pointed at the same gap: as AI systems take on more work, the tools for defending them have not kept pace. Investors are betting that the company that closes that gap first will be worth a great deal.

Fortune 500 adoption is part of the story too. Corma says its models are already deployed at large organizations across healthcare, finance, energy, critical infrastructure and retail.

The company also says it has expanded its server and compute capacity 24-fold since August. For a startup whose biggest constraint is reportedly computing power, that expansion is as important as the funding itself.

The Numbers Behind the Hype

Corma’s own performance claims are eye-catching. The company reports that 99 percent of security tasks are completed autonomously and successfully, with end-to-end completion 100 times faster and cost per completed task 95 percent lower.

One anonymized Fortune 1000 services company, quoted on Corma’s product page, said its end-to-end response time fell 94 percent compared with its former service provider, which it subsequently replaced with Corma.

Those figures have not been independently audited, and they come from the company itself. That does not make them false, but it does mean they should be read as marketing claims until outside verification arrives.

The same caution applies to the valuation. A $1 billion price tag for a company with about 25 employees and a product that is months old reflects investor belief in a future, not revenue today.

Still, the backers are not naive. NEA, Sequoia, Khosla and Coatue have all funded cybersecurity winners before, and all four chose to put more money into Corma rather than wait for proof.

A Crowded Field Gets More Crowded

Corma is not entering an empty market. Established security vendors are adding AI agents to their products, and a wave of startups is chasing the same overwhelmed security teams.

What sets Corma apart, at least on paper, is the depth of its bet. Most competitors fine-tune existing models for security tasks. Corma is building domain-specific foundation models, rethinking architecture, extending context windows to catch faint anomaly signals, and using adversarial self-play to harden its systems.

The company says its development program also includes continuous learning that can establish baselines over months and years, so the models understand what normal looks like for each customer before flagging what is not.

That approach is expensive. Training foundation models requires enormous compute, which is why Corma says the new funding will go toward scaling its models, expanding compute capacity and selective hiring.

The next generation of its foundation models is coming soon, according to Pluda and his team, though no release date has been given. How those models perform in independent testing will matter far more than any funding announcement.

What Happens Next for Corma

The immediate challenge is not money but capacity. Corma says demand from the world’s largest organizations has been overwhelming, and its ability to expand access depends on obtaining enough infrastructure to serve both current customers and organizations on its waitlist.

That is a good problem for a startup to have, but it is still a problem. Every customer waiting for access is a customer a competitor can win.

The longer question is whether defense-specific models really outperform general-purpose ones in the real world. Corma’s argument is elegant, but the security industry has heard elegant arguments before.

For now, though, the market has spoken. A startup that did not exist in the public eye three months ago is now a unicorn, backed by four of the most powerful names in venture capital.

Whether Corma becomes the next great cybersecurity company or a cautionary tale about AI funding excess will be decided by its models, not its valuation. The security teams waiting on its waitlist will be watching closely.

Source: TechCrunch

Related reading: Fireworks AI Breach Widens: Customer Datasets and Model Weights Exposed in Second Hack

Related reading: Railway Breach: Stolen AI Keys Rack Up $740,000 in Fraud Charges

Related reading: Microsoft’s Lazarus Backdoor: Fewer Than 10 Neurons Can Turn an AI Agent Rogue

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top