Corma Raises $125M Series A at a $1B Valuation: What Security Founders Should Know

Corma, the AI security startup with offices in San Francisco and Tel Aviv, announced a $125 million Series A on Monday, November 16, 2026. SiliconANGLE reports that NEA led the round, that returning backers Khosla Ventures, Coatue and Sequoia Capital joined, and that the company is now valued at $1 billion. For a company that came out of stealth in August, that is a fast climb, and it tells you a lot about where security budgets and investor attention are heading right now.

If you build, sell to or run security teams in the Bay Area, this round is worth a close read. Corma is betting that general-purpose AI models help attackers more than defenders, and that defense needs its own foundation models trained on the messy data security teams live in every day. Below is what the company says it builds, what the money is for, and what the deal signals for founders and operators working near cybersecurity.

What Corma builds and why defense needs its own models

Corma describes itself on its website as training foundation models for defensive cybersecurity, deployed as an AI security workforce that works alongside human teams. The pitch rests on an asymmetry. Offense lives in code, exploitation and social engineering, which is exactly what popular coding models are good at, while defense lives in audit logs, network flows and security events, where a single missed signal can matter.

According to SiliconANGLE, Corma ran a test in which several large language models were asked to hack a simulated corporate network. The models installed a malicious file in 85% of test runs but found only 12% of the malicious files they had installed themselves. Corma uses that gap to argue that general models give attackers an edge, and it says its own model outperformed several leading LLMs on a benchmark called DefenseBench V5, which measures how reliably and how fast a model completes defensive tasks. Those results come from the company, so treat them as claims until outside teams publish their own numbers.

How the product works inside a security team

The day-to-day product ingests telemetry from tools a security team already runs, including firewalls and network monitoring software, then looks for breaches and writes remediation guidance for each threat it finds. Corma says its models can run breach investigations on a fully autonomous basis and alert the right person when human input is needed. Teams can also steer the system through a Slack chatbot, for example by asking it to confirm that a malware cleanup on a compromised server actually worked.

The hard part, as SiliconANGLE notes, is that breach response is full of false positives and duplicate alerts, and models often see only partial information because many companies discard security logs quickly to save on storage. That detail is useful for anyone selling into security operations. If your product generates or stores telemetry, longer retention and cleaner data may become more valuable as AI investigators move into the stack.

Where the $125 million goes

Both SiliconANGLE and PYMNTS say the money is mostly about capacity. Corma plans to buy more infrastructure and improve its models, and PYMNTS adds that the round will help the company expand staffing and compute. In its own announcement, quoted by PYMNTS, Corma said that securing enough compute to keep up with demand is one of its biggest ongoing challenges and that it is still working through a waitlist.

The speed of the raise stands out. SecurityWeek reported in August that Corma emerged from stealth with $60 million in seed funding from Sequoia Capital, Khosla Ventures and Coatue, and that the company was founded in 2025 with Alon Pluda as CEO. About 3 months later, the same backers came back alongside a new lead. PYMNTS also notes that Corma tied the surge in demand to a wave of critical vulnerabilities and breaches in October, a stretch Corma says the industry called "Black October."

What the round means for Bay Area founders and operators

For security founders, the clearest takeaway is that investors are now funding model companies built for a single domain, in addition to applications on top of general models. If you are building in detection, response or compliance, expect buyers to ask how your product compares with a security-specific model, and be ready with your own evaluation data. Our recent look at Ballistic Ventures' $490M cybersecurity funds covers more of the investor appetite in this space.

For operators, the practical question is workflow. A product that investigates alerts on its own and reports back through Slack changes what a small security team can handle, and it raises new questions about review, audit trails and who signs off on remediation. Agent tooling is moving in the same direction across the stack, which we covered in our post on LangChain's $3B Series C.

For tech workers, compute is the bottleneck to watch. When a company says demand outruns capacity 3 months after launch, hiring tends to follow in infrastructure, model training and customer deployment roles. Corma also joins a growing list of young AI companies valued at $1 billion or more, a trend we tracked in our October 2026 Bay Area unicorns roundup, and it sits alongside new San Francisco research labs like the ones in our post on AI neolabs in SF. On the robotics side, our Rhoda AI explainer looks at how one robotics startup is training robots on internet video.

Questions worth asking before you build or buy

If you are evaluating an AI security product, ask which data sources it reads, how long it needs logs retained, and how it handles a finding it cannot confirm. Ask who can trigger remediation and whether every action leaves a record a human can review later. Those questions apply to Corma and to any competitor, and the answers will shape whether a tool reduces work or adds a new queue.

If you are building in the space, the Corma story suggests that a credible benchmark and a clear data story can carry a lot of weight with investors. Publishing how you test, and inviting customers to test on their own data, is one way to stand out while the market is still sorting out whose claims hold up.

Find your people in AI and security

Funding news moves fast, and the best context often comes from people who work in security every day. If you want to trade notes on AI models, agents and the startups behind them, join the AI Trends Tribe, where founders, engineers and curious operators talk through stories like this one. You can also meet the Bay Area tech crowd in person through the Tribe Events calendar on Luma, which lists upcoming talks, dinners and meetups, including the AI conversations Tribe hosts on the Peninsula.

Previous
Previous

Rhoda AI Explained: How a Palo Alto Startup Trains Robots on Internet Video

Next
Next

Roommate Group Chat Names: 70 Ideas for Your Shared Place