Corma announced a $125 million Series A on November 16, 2026, to expand its defensive cybersecurity AI models and computing capacity. The company, with operations in Tel Aviv and San Francisco, is pursuing autonomous security agents: a development relevant to universities evaluating AI systems that can act inside institutional networks. Corma announced the financing; Globes reported its operating locations.
NEA leads a new round after August’s seed announcement
The Series A values Corma at $1 billion and brings its total funding to $185 million. NEA led the round, with returning investors Sequoia Capital, Khosla Ventures and Coatue participating, according to the company announcement. This is a new financing, separate from the $60 million seed round disclosed three months earlier.
Corma’s dated announcement index places its emergence from stealth on August 10 and the Series A on November 16. That sequence matters: the latest announcement adds capital to an existing defensive AI effort rather than introducing the company for the first time.
| Announcement | Date | Funding |
|---|---|---|
| Emergence from stealth and seed round | August 10, 2026 | $60 million |
| Series A | November 16, 2026 | $125 million |
Both rounds appear in Corma’s announcement history. The $1 billion figure is the valuation attached to the latest round, while $185 million is cumulative financing; those figures describe different aspects of the transaction.
Why Corma is building models specifically for defense
In an interview with Globes, CEO Alon Pluda argued that general-purpose language models’ strengths in software development and vulnerability discovery do not automatically equip them for defense. He described defensive work as finding significant signals in enterprise telemetry, logs and events. This is Corma’s explanation of its technical approach, rather than an independent finding about every competing model.
Globes reported that Corma develops language models from scratch and builds agents that work alongside customers’ cybersecurity teams. The distinction is between training a model for a specialized operational task and adding a conversational interface to existing software. For researchers studying agent behavior, that makes the evaluation target the model’s performance within a security workflow, including its use of tools and interpretation of events.
An institutional model evaluation suite offers a useful way to organize the evidence needed to assess such a system. CASRAI’s interpretation is that coding performance alone would be an incomplete basis for judging the defensive workflow Pluda describes.
The CEO describes agents working through existing security tools
In a separate interview reported by CTech’s Meir Orbach, Pluda described Corma’s offering as an AI workforce integrated with an organization’s existing security stack. He said its agents can carry out security tasks autonomously, without human intervention. That is a company capability claim, and it should be read as such when assessing how much operational responsibility to delegate.
CTech reported that the company serves large enterprises in healthcare, financial services, energy and critical infrastructure, retail, and telecommunications. The outlet also relayed Corma’s claim that it had expanded server capacity 24-fold following demand from major enterprises. That expansion describes infrastructure growth; it does not measure detection accuracy, false alarms or the safety of autonomous actions.
Pluda’s comments frame the investment as a race to improve defensive capabilities. For research groups examining the announcement, the useful distinction is between investor backing, customer demand and demonstrated task performance. Each addresses a different question about the company and its systems.
What the development means for universities and research offices
The governance context is broader than this funding round. NIST’s AI Risk Management Framework is voluntary guidance intended to incorporate trustworthiness into the design, development, use and evaluation of AI products, services and systems. NIST also describes its generative AI profile as a resource for identifying risks and selecting actions aligned with an organization’s goals and priorities.
Applied to Corma’s stated approach, CASRAI’s analysis is that a university evaluating autonomous defense would need to connect technical testing with decisions about operational authority. Relevant evaluation questions include which security tools an agent can use, what actions it may take, how its activity can be inspected, and when staff can interrupt or escalate its work. These are institutional assessment questions, rather than features established by the financing announcement.
Research offices and campus security teams can use an AI risk assessment and risk register to connect an intended use with the evidence and responsible decision makers. The NIST AI framework’s application to research institutions provides further background. The immediate relevance is the prospect of delegating security operations to agents, which makes the scope of that delegation a central evaluation issue.
Compute expansion and further models are next
Corma says the funding will support model scaling, additional compute and selective hiring. It describes securing sufficient compute as a continuing constraint and says it is expanding capacity for organizations on its waitlist. The company also says it will share its next generation of foundation models soon, without setting a release date in the announcement.
Sources
- Corma, “Three months in”, November 16, 2026.
- CTech, Meir Orbach, “Cyber startup Corma goes from stealth to unicorn in just three months with $125 million Series A”, November 16, 2026.
- Globes, Assaf Gilead, “Defensive AI agents co Corma raises $125m at $1b valuation”, November 16, 2026.
- Corma announcement index, entries dated August 10 and November 16, 2026.
- NIST, AI Risk Management Framework overview, framework released January 26, 2023; generative AI profile released July 26, 2024.
Sources read November 16, 2026.







