AI security startup Corma hits a $1bn valuation three months after launch

4 Sources

Share

Corma has raised a $125m Series A at a $1bn valuation, three months after leaving stealth. The company ties the demand to October's breach wave, but the revenue and benchmark evidence are largely its own.

Corma, a startup about three months out of stealth that trains its own AI models for cybersecurity defence, has raised a $125 million Series A at a $1 billion valuation

2

4

1

3

. The round was led by New Enterprise Associates, with Sequoia Capital, Khosla Ventures and Coatue returning

2

.

Source: SiliconANGLE

Source: SiliconANGLE

Black October and a waitlist

The company ties the demand spike to October's wave of critical software flaws and breaches, a stretch the industry dubbed Black October

2

4

1

. It included a critical flaw in the PHP programming language disclosed on October 21, alongside Node.js and LiteLLM incidents, and drew warnings from governments from Washington to Tokyo

2

4

1

. Corma says demand from the biggest companies, on the Fortune 100, 500 and Global 500 lists, has been overwhelming since it emerged in August

2

4

3

. It grew server capacity 24-fold to keep up, and runs a waitlist while it adds compute

2

4

3

.

"Since we came out of stealth, demand from the world's largest organizations has been overwhelming, and one of our hardest problems has been securing capacity fast enough to keep up with the demand," said co-founder and CEO Alon Pluda, an alumnus of Israel's Unit 8200 intelligence unit

2

3

.

Source: The Next Web

Source: The Next Web

The August exit from stealth, about three months ago, came with a $60 million seed round led by Sequoia, so total funding now stands at $185 million

2

4

1

. The new money is earmarked for scaling the models, buying more compute and selective hiring, which the company says it needs to meet overwhelming demand and clear its waitlist

2

4

3

. Pluda has promised the next model generation soon, without a date

2

3

.

What the numbers do and do not say

No revenue figure was disclosed; Pluda told Calcalist any figure he gave might already be outdated

2

. Per that reporting, customers work in healthcare, financial services, energy, critical infrastructure, retail and telecommunications

2

.

The performance evidence is thinner. In a test the company ran itself, several large language models installed a malicious file in 85% of simulated corporate-network hack attempts

3

. They found only 12% of the files they themselves installed

3

. Its own benchmark chart, labelled DefenseBench V5, is self-run, and puts Corma v0.4 ahead of Claude, Gemini, GPT-6-family, Kimi and Grok models on defensive task performance and execution speed

2

3

. Neither has been independently checked

2

3

.

The bet on defence

Corma's own thesis is that AI models improve at offence faster than defence

2

4

1

. It trains its own models to sift audit logs, security alerts and network traffic hunting weak signals, then sells them as agents that operate a company's existing security tools

2

4

1

3

. The company says its models can speed some breach-response workflows a hundredfold while cutting costs, and can run breach investigations on their own, alerting humans only when needed

2

3

. Administrators can also steer them through a Slack chatbot

2

3

.

Source: TechCrunch

Source: TechCrunch

The job is genuinely hard, per SiliconANGLE's reporting

3

. Models must filter out floods of false positives and duplicate alerts, they often see only part of a breach, and many companies delete security logs quickly to save on storage

3

. "Cybersecurity is now a battle of intelligence against intelligence, and Black October made that impossible to ignore," Pluda said

2

. The company had warned at its August launch that AI offence was improving faster than defence, and says that warning played out in public during October's incidents

4

. Its homepage also accuses unnamed AI companies of profiting from offensive capabilities while preaching safety

2

. All of this sits inside a company of about 25 people with offices in San Francisco and Tel Aviv

2

3

.

A crowded field, one investor disagrees

Aaron Jacobson, a partner at the firm that led the round, said: "Corma is the only company taking this on in a way that is scalable across organizations to prevent the next big cyberattack"

2

. TechCrunch's reporting points the other way, describing a field crowded with AI defence work

1

. Per TechCrunch, that field includes Kevin Mandia's new startup Armadin, Anthropic, OpenAI and Microsoft

1

.

© 2026 TheOutpost.AI All rights reserved