Corma’s $125 million Series A brings total funding to $185 million as the defensive cybersecurity AI lab confronts a computing bottleneck. Its specialized-model strategy faces existing security agents and an unresolved test of performance and deployment costs.
Corma, an Israeli AI lab founded in 2025 to develop its own foundation models for defensive cybersecurity, announced a $125 million Series A on November 16, 2026, at a $1 billion valuation. Three months after its public debut, the company is raising money to expand both the models behind its autonomous security workforce and the computing capacity needed to serve enterprises.

Corma's company-reported financing: $60 million at its public debut and a subsequent $125 million Series A. Bars represent individual rounds. Source: Corma.
Venture investor NEA led the round, with existing investors Sequoia Capital, Khosla Ventures and Coatue participating, Corma said in its funding announcement. It emerged from stealth in August with $60 million; the Series A brings total financing to $185 million.
Corma says demand from large organizations has outstripped its ability to secure computing capacity, leaving prospective users on a waitlist. It plans to scale its foundation models, increase compute capacity and selectively expand its team. The financing therefore addresses a delivery constraint as well as further research.
Co-founder and chief executive Alon Pluda previously co-founded cybersecurity company Miggo Security and held technical roles in Israel’s military intelligence Unit 8200. A financing report says he left Miggo following a dispute with his co-founders. He founded Corma in 2025.
The company operates in Tel Aviv and San Francisco. Its team includes researchers from Google and DeepMind and cybersecurity specialists from Unit 8200, according to the accompanying coverage. That combines model-development expertise with security experience, although the company remains small: Globes reports 20 employees, while CTech puts the figure at approximately 25.
Corma develops language models from scratch for defensive security rather than relying on existing general-purpose models, the financing report says. Its product, the Corma Workforce, uses an organization’s existing security tools to investigate threats and perform defensive tasks alongside its security team.
Pluda argues that improvements in coding help general-purpose models find vulnerabilities but do not necessarily equip them to interpret logs, telemetry and security events consistently. In the accompanying interview, he claimed Corma’s models were an order of magnitude better and described its workflows as fully autonomous, without human intervention. The coverage supplies no defined benchmark for that performance claim or independent verification of the autonomy claim.
There are already alternatives offering defensive automation. Microsoft’s Security Copilot, an AI assistant for detecting, investigating and responding to security incidents, had launched a year before Microsoft’s March 2025 agent announcement. That announcement set out six Microsoft-built and five partner-built agents for preview in April 2025.
Microsoft described agents for phishing and data-security alert triage, identity-policy gaps, vulnerability remediation and threat intelligence. Its design retained security-team control, including administrator approval for Windows patches. These are vendor-described capabilities, rather than a comparative test against Corma. They nevertheless narrow Corma’s novelty claim: buyers already had an agent-based approach to defensive work to evaluate. Corma must demonstrate what its specialized models add.

Microsoft's March 2025 diagram shows its company-reported agent design, including human supervision and continuous learning. It provides no performance comparison with Corma. Source: Microsoft.
In an August research post, Corma tested four general-purpose frontier models as attackers and defenders in a synthetic enterprise network with no real users or data. Both sides had root-equivalent access and a 100-turn budget. This tested planting and finding persistence after compromise, rather than whether an attacker could first break into a protected enterprise.
Corma reports verified, reboot-surviving backdoors in 85% of engagements and implant detection in 19% across all engagements. Among runs with confirmed persistent implants, detection was 22%. These measures have different denominators and scoring rules; they are not customer breach rates.
The post reports 241 engagements, although four models in every attacker–defender pairing, with 15 runs per pairing, implies 240. It does not reconcile that discrepancy. Its proposed explanation—that general-purpose models’ training better suits offense—is explicitly an interpretation. The experiment supports scrutiny of autonomous defensive reliability, but supplies no comparison demonstrating Corma’s own models’ advantage.

Corma's company-reported implant detection rates in a synthetic network. Detection required the correct host plus mechanism or location. These are not customer breach rates or a test of Corma's models; the source's engagement count is inconsistent. Source: Corma research.
Corma says it increased server capacity 24-fold after its August unveiling to meet demand from Fortune 100, Fortune 500 and Fortune Global 500 organizations, a company statement reproduced in the financing coverage. That multiplier has no disclosed starting server count or computing expenditure, so it cannot establish the absolute scale or efficiency of deployment.
The reports describe customers across healthcare, financial services, energy and critical infrastructure, retail, and telecommunications. The financing report estimates annual recurring revenue in the tens of millions of dollars, without a precise figure or explanation of the estimate’s basis. It is an estimate, distinct from Pluda’s statement that revenue is growing rapidly.
The disclosed material gives neither pricing nor unit costs, and does not establish the number of paying customers. Those gaps prevent a useful valuation-to-revenue calculation or an assessment of how much revenue remains after model development and computing expenses. More capacity can support more work; the unresolved commercial question is how much reliable defensive work each dollar buys.
Corma says it will share its next generation of foundation models soon, without a release date in the announcement. Its goal of defensive cybersecurity superintelligence remains an ambition.
For enterprise buyers, the next decision is whether to delegate more security work to these systems. Comparable detection and false-positive results, the boundaries of autonomous action, required human oversight and the cost of completed tasks would help them assess that decision against existing tools. The funding gives Corma resources to expand; the next model release and evidence from customer deployments must show whether its specialized approach warrants that delegation.
Get concise AI news and useful context from the Magica team.
Read the newsletterManufacturers, utilities and miners linked to data center construction have shed $4.2 trillion in market value since June. Strong supplier orders contrast with financing concerns, project delays and political resistance.
RIKEN has stopped Fugaku’s research jobs after an intrusion into its operational network. Up to 4,597 users’ account information may be affected, and intruders tried acquired credentials on other systems; research-result leakage remains unconfirmed.
DeepSeek users report daily caps, shorter inputs and restrictions on resuming inactive conversations. Reported financing and a planned data center do not establish when more capacity will reach the free chat service, whose quota rules remain unclear.
Google found an attacker framework configured to compare software patches, develop exploits and execute them without human intervention. The finding sharpens the pressure on defenders to patch quickly, but does not establish that entire attacks now run autonomously.
Ask Maps is adding shareable travel schedules, saved-list availability searches and U.S. appointment finding. The update extends an existing AI planning service rather than introducing trip planning to Maps.
SemiAnalysis has reportedly cut its 2030 global AI accelerator spending forecast by about 40% to $923 billion, with Nvidia facing the largest revisions. Lower price assumptions and financing constraints help explain the reduction, while customer deals show both the strength and financial cost of competing chip platforms.
Federal Reserve Governor Michael S. Barr says banks report AI agents attempting account actions without verifiable permission. Autonomous deposit transfers could accelerate future bank runs, but their scale and speed remain unmeasured in his speech.
Amazon-owned Zoox is opening its Las Vegas robotaxis to Uber riders while keeping its own app. The November launch follows an earlier summer target; a joint MGM Resorts agreement addresses resort pickups, while fleet size and financial terms remain undisclosed.
Germany’s reported military AI shortlist excludes Palantir, while France and the Netherlands pursue European alternatives. The challenge is to secure national control over data and infrastructure while delivering usable systems that work with allies already adopting the US supplier.
China aims to make school AI education broadly available by 2030. National guidance already calls for pupil assessment, but an investigation finds common measures of AI literacy still lag an uneven rollout spanning courses, tutoring and student monitoring.
Drive Capital led Zus Health's $33 million Series B, extending a shared-record strategy backed by earlier financing and EHR partnerships. The company is expanding AI workflows and pursuing a more direct role in national data exchange, while disclosed activity figures leave customer returns unmeasured.
Atomarine’s $19 million seed round and Mocean Energy’s wave-powered computing proposal target land, water and electricity constraints. Earlier underwater projects show technical promise, but do not establish the costs or performance of these new floating designs.
Tencent’s new Hunyuan model charges 2 yuan per million output tokens and targets everyday office work. Its claimed 95% task-cost saving against Hy4 comes from an internal benchmark, with a trade-off on long coding tasks.
The Indian chip startup is funding programmable image sensors and targeting production in 2027. Its promised system savings remain unproven, while Sony's earlier designs challenge broad claims of novelty.
Baidu’s AI cloud infrastructure revenue nearly doubled in the third quarter, cushioning a 22% advertising decline. But AI applications were flat, app users fell, and the return to reported profit obscured lower adjusted earnings and a wider annual free-cash-flow deficit.
The Trump administration reportedly plans to designate more than two dozen AI companies to help win overseas business. Existing export rules support integrated technology packages, but guarantee neither financing nor licenses and leave room to support competing suppliers.
Singular leads the Danish accounting software company's seed round as it builds a London commercial hub and targets Benelux. Its pitch rests on automating preparation while accountants retain sign-off, with claimed efficiency gains still awaiting independent verification.
Alibaba says Qwen4-Max reduces average task costs by 62% on its workplace benchmark. The hosted launch builds on its predecessor’s agent capabilities as Alibaba expands a growing cloud business and bears the expense of AI infrastructure.
The FBI warns that ARTEX can accelerate intrusions, with some reaching privileged access in under three hours. Microsoft also documents unsuccessful mass targeting, while compromised backup-service credentials opened a separate route into customer environments.
Jensen Huang has reportedly discussed returning to Nvidia three days after resigning. Supporters want reinstatement during an investigation, while directors have installed interim leadership and begun a permanent CEO search.