LaunchedEditorial Listing

Corma

Corma · Corma: AI Security Workforce That Runs on Its Own Defense-Trained Foundation Models

Open Corma

Corma is an AI security workforce for large enterprise security teams: agents running on Corma's own foundation models, trained for defensive cybersecurity, take on security tasks such as investigations, threat hunts, detection engineering and containment from start to finish, working through the SIEM, EDR, identity, cloud and firewall tools a team already runs and taking requests, questions and action approvals in Slack or Microsoft Teams. Access is by request, with a waitlist, and there is no public pricing or documentation.

PricingCustom
Setuphard
Open sourceNo
DocsNo
CategorySecurity
Security OperationsSOCThreat HuntingThreat IntelligenceIncident ResponseDetection EngineeringAttack Surface ManagementVulnerability ManagementSIEMFoundation ModelsHuman-in-the-LoopSlackMicrosoft TeamsEnterprise

Best for

Large enterprise security teams (CISOs and SOC, detection engineering, cloud security and IAM leads) in sectors such as healthcare, finance, energy, critical infrastructure and retail that want AI agents to take whole security tasks off their queue inside their existing stack, with humans approving response actions

Not ideal for

Small and mid-sized companies, teams that need to start immediately or evaluate from public pricing and documentation, organizations that require a documented self-hosted or on-premises option, and buyers who need independently verified performance data before shortlisting a vendor

Who it's for

CISOs, security operations, detection engineering, cloud security and identity teams at large enterprises

Capabilities

  • Agents powered by Corma's own foundation models, which the company trains specifically for defensive cybersecurity work such as analyzing logs, events, configurations and network flows
  • Takes security tasks end to end: works out next steps, moves between tools to investigate and connect evidence, takes action and brings in a person when needed
  • Covers security roles beyond alert triage: cloud security engineering, detection engineering, threat intelligence, digital forensic analysis, identity and access management, and attack surface management
  • Works in the team's existing identity, cloud, endpoint, SIEM and firewall/IDS products; tools pictured on the workforce page include Splunk, Microsoft Sentinel, Google Chronicle, IBM QRadar, Elastic, CrowdStrike, SentinelOne, Microsoft Defender, Okta, Microsoft Entra ID, Ping, CyberArk, AWS, Azure, Google Cloud, Wiz, Orca, Palo Alto Networks, Fortinet, Cisco and Zscaler
  • Chat in Slack or Microsoft Teams to assign tasks, ask questions, review findings and approve actions, as with a human teammate
  • Containment actions such as endpoint or server isolation, with approval requests sent to the security team; the site's examples include both approval-gated isolation and cases shown already contained, and it does not document which actions require approval
  • Example workloads on the site include threat-intelligence hunts (for instance for fraudulent DPRK IT-worker identities), DLP alerts resolved against policy, exposure and exploitability analysis for a newly disclosed critical RCE, and suspected data exfiltration to a personal email account
  • Compliance badges for AICPA SOC, ISO 27001, GDPR and HIPAA are displayed on the workforce page

Limitations

  • Access is by request only: the Series A announcement mentions a waitlist and says securing enough compute capacity to meet demand is one of the company's biggest challenges
  • Pitched at Fortune 500-scale enterprises; the request-access form takes companies under 1,000 employees, but nothing on the site addresses smaller teams
  • No public pricing, documentation, integration list, API reference or deployment details; the tools shown on the site are logos in an illustration, not a documented connector catalog
  • No self-hosted or on-premises deployment is documented, and the site does not say where customer security data is processed or whether it is used to train Corma's models
  • Which actions run without human approval is not documented: some examples end with an endpoint isolated or a case closed with no approval step shown
  • Performance figures (99% of tasks completed autonomously, 100x faster, 10x more work per team, 95% lower cost per task), the DefenseBench V5 chart on its homepage and the claim that its model has far surpassed every frontier model on defensive tasks are vendor-reported without published methodology
  • Customer evidence on the site consists of anonymized CISO quotes; no named customers or case studies are published
  • Young company: it emerged from stealth in August 2026, so its track record in production is short

Use cases

  • Handing threat-intelligence leads, such as signs of fraudulent remote IT-worker identities, to an agent that hunts across identity and endpoint data and reports back
  • Investigating high-severity DLP and data-exfiltration alerts, closing benign ones and containing the endpoint when exfiltration is confirmed
  • Working out an organization's exposure and exploitability when a critical vulnerability is disclosed
  • Evaluating detection coverage for new attack techniques as part of detection engineering
  • Hunting for infections that existing tools missed, such as ClickFix-style attacks, and containing affected machines
  • Adding capacity to a stretched enterprise security team, or running alongside an outsourced SOC provider to compare results

Our take

Corma is a bet on a specific thesis: that defending a network is a different skill from writing code, and that a model trained for defense will outperform agents wrapped around general-purpose LLMs. If that holds, a workforce that can be handed whole tasks, from a threat-intel hunt to an exposure check to a contained endpoint, is a bigger step than another alert-triage layer, and its Slack- and Teams-based approval flow fits how security teams already work. The difficulty for a buyer is that almost everything that would confirm the thesis is not public yet: there is no documentation, pricing, integration catalog or deployment detail, the performance numbers are Corma's own, and access runs through a waitlist constrained by compute. For a large enterprise security team with the leverage to run a proof of concept alongside its current SOC tooling or service provider, Corma is worth a request. Teams that need to evaluate from documentation, start this quarter, or are below enterprise scale should look at the more established AI SOC agents first.

Who should use it

CISOs and security leaders at large enterprises, especially in regulated or critical sectors, who want to delegate whole security tasks (investigations, hunts, detection coverage, identity and cloud security work) to AI agents inside their existing stack, and who can run a proof of concept while waiting for access.

Who should skip it

Small and mid-sized organizations, teams that need self-serve access or published pricing, organizations that require a documented on-premises or self-hosted option, and buyers who need public documentation or independently verified benchmarks before evaluating a product.

Strengths

  • Runs on foundation models trained specifically for defensive security rather than a general-purpose LLM
  • Covers a wider range of security roles than alert-triage agents, including detection engineering, cloud security, IAM, forensics and attack surface management
  • Works inside the tools a team already has and talks to the team in Slack or Microsoft Teams
  • Can carry a case through to containment, with approval requests to the team in Slack or Teams
  • Well funded, with backing from NEA, Sequoia Capital, Khosla Ventures and Coatue

Weaknesses

  • Request access with a waitlist; capacity is limited by compute
  • No public pricing, documentation, API or integration list to evaluate before talking to the company
  • Headline performance numbers are vendor-reported, without published methodology
  • No documentation of which actions need human approval or how autonomy is configured
  • Aimed at large enterprises, so smaller teams are unlikely to be a fit
  • Very young company with a short public track record

Corma pricing

Enterprise

Contact sales

  • Access through a request-access form with a waitlist
  • AI security workforce running on Corma's foundation models
  • Works with the existing security stack and in Slack or Microsoft Teams

Note: Corma does not publish pricing. Access is requested through a form on corma.ai that asks for company size and role, and the company says it is working through a waitlist while it expands compute capacity.

Technical specs

Available models

Corma's proprietary foundation models for defensive cybersecurity

Where Corma excels

Responding to an active intrusion outside working hours

Corma investigates across endpoint, identity and SIEM data, prepares containment, and asks for approval in Slack or Teams, so an on-call lead can approve isolation from a phone and let the agent finish the job.

Checking exposure when a critical vulnerability drops

Instead of an analyst manually querying asset, cloud and endpoint tools, the workforce can investigate which systems are exposed and whether the flaw is exploitable in the environment.

Insider-risk and DLP investigations

Corma's examples include a DLP alert for an SSN copied to USB that it resolves as benign personal activity, and a confirmed exfiltration of company files to a personal email account that ends with the employee's endpoint isolated and HR and Security notified.

Stretching an enterprise security team

Teams that cannot hire fast enough can hand detection engineering, threat-intel hunts and cloud or identity tasks to the workforce and keep analysts on judgment calls.

Corma vs. competitors

Corma vs. Dropzone AI

Dropzone AI is an AI SOC analyst and threat hunter that investigates alerts across your SIEM, EDR, identity and cloud tools, built on third-party LLMs, with public docs, a REST API and a documented integration list. Corma trains its own defense-specific models and pitches a broader workforce covering detection engineering, cloud security, IAM and forensics, but publishes much less detail and is request-access only.

Corma vs. Purple AI

Purple AI is SentinelOne's agentic security analyst, sold as part of the SentinelOne Singularity Platform and working over data in SentinelOne's data lake. Corma is not tied to one platform: it says it works across the SIEM, EDR, identity and cloud tools a team already runs, with SentinelOne among the tools pictured on its site.

Corma vs. Mave

Mave is an agentic security operations platform that investigates alerts, hunts, tunes detections and runs response actions on top of a team's existing tools, with destructive actions held behind approval by default. Neither publishes documentation or pricing. Corma differs in training its own foundation models and in pitching a workforce that takes on broader roles such as cloud security, IAM and forensics, while Mave states its approval defaults more explicitly.

Corma vs. Brava

Brava focuses on detection engineering: it models attacks, synthesizes the telemetry they would leave and writes missing SIEM detections. Corma lists detection engineering as one of several roles its workforce covers, alongside investigations, threat intelligence, cloud security and identity work.

Frequently asked questions

What is Corma?

Corma is a company training foundation models for defensive cybersecurity and deploying them as the Corma Workforce, AI agents that work alongside an enterprise security team. The agents take on tasks such as investigations, threat hunts, detection engineering, cloud security and identity work, use the team's existing security tools, and ask a person to approve actions through Slack or Microsoft Teams.

How much does Corma cost?

Corma does not publish pricing. Access is requested through a form on its website, and the company says there is a waitlist while it expands compute capacity, so pricing is agreed directly with Corma.

Who can use Corma?

Corma is aimed at large organizations. Its workforce page says it is trusted by the Fortune 500 and live across healthcare, finance, energy, critical infrastructure and retail. Access is requested through a form that asks for work email, company, role and company size (bands from under 1,000 to 50,000+ employees), and the company says it is working through a waitlist. There is no self-serve sign-up.

Does Corma take actions on its own?

Corma says its workforce handles tasks end to end, including taking action, and brings in a person when needed; team members can approve actions from Slack or Microsoft Teams. The examples on its site go both ways: one Slack thread shows Corma asking for approval before isolating a laptop and a vulnerability case waits for approval to isolate a server, while other examples end with an endpoint already isolated or a DLP case closed automatically. The site does not document which actions require approval or how that is configured.

Which security tools does Corma work with?

Corma says it works with a team's existing identity, cloud, endpoint, SIEM and firewall/IDS products. Its workforce page pictures tools including Splunk, Microsoft Sentinel, Google Chronicle, IBM QRadar, Elastic, CrowdStrike, SentinelOne, Microsoft Defender, Okta, Entra ID, CyberArk, AWS, Azure, Google Cloud, Wiz, Palo Alto Networks, Fortinet and Zscaler, but no documented integration list is published.

Corma vs Dropzone AI: what is the difference?

Dropzone AI is an AI SOC analyst and threat hunter that investigates alerts from your SIEM, EDR and other tools, with public documentation, a REST API and a documented integration catalog. Corma trains its own foundation models for defense and positions its workforce more broadly, across detection engineering, cloud security, IAM, forensics and attack surface management, but it is request-access only and publishes far less technical detail.

Integrations & fit

SlackMicrosoft TeamsSplunkMicrosoft SentinelGoogle ChronicleIBM QRadarElasticCrowdStrikeSentinelOneMicrosoft DefenderOktaMicrosoft Entra IDPing IdentityCyberArkAWSMicrosoft AzureGoogle CloudWizOrca SecurityPalo Alto NetworksFortinetCiscoZscaler
Good fit forEnterprise
Pricing modelCustom· Contact for pricing
See pricing on Corma →

Alternatives to consider

About Corma

Corma describes itself as a frontier AI lab for defensive cybersecurity rather than a SOC tool vendor. Its argument, set out in its launch post and an August 2026 research write-up, is that general-purpose models are improving much faster at offense (code, exploitation) than at defense (audit logs, network flows, security events), so it trains its own domain-specific foundation models for the defensive side. Those models are delivered as the Corma Workforce: agents that sit alongside an enterprise security team, use the same identity, cloud, endpoint, SIEM and firewall/IDS products the team already uses, and carry a task through to completion, bringing a person in when judgment or approval is needed. The roles Corma lists are cloud security engineering, detection engineering, threat intelligence, digital forensic analysis, identity and access management, and attack surface management, and the worked examples on its site include hunting for DPRK IT-worker identities, investigating a DLP alert for SSNs copied to USB storage, assessing exposure to a newly disclosed critical RCE, and running down a suspected data exfiltration. Teams talk to the workforce in Slack or Microsoft Teams to assign tasks, ask questions, review findings and approve actions. The examples show a mix of approval-gated and completed actions: in one Slack thread Corma hunts for missed ClickFix infections, asks for approval and then isolates the affected laptop; a vulnerability-exposure case stops at "server isolation prepared, approval requested"; the identity-fraud and exfiltration cases end with the endpoint already isolated and HR notified; and a DLP alert is closed automatically as benign. Corma does not document which actions need approval. What sets Corma apart from SOC agents built on third-party LLMs is the in-house model and the broader set of security roles beyond alert triage. The tradeoffs are about access and verifiability. Corma pitches itself to the largest organizations (its workforce page leads with "Trusted by the Fortune 500"), its Series A post says there is a waitlist and that compute capacity is a constraint, and there is no public pricing, documentation, integration list, API or deployment detail. Its headline figures (99% of tasks completed autonomously, 100x faster completion, 95% lower cost per task), the DefenseBench V5 chart on its homepage and its claim that its model has far surpassed every frontier model on defensive tasks are the company's own, without published methodology. The company raised a $60 million seed in August 2026 and a $125 million Series A at a $1 billion valuation in November 2026.

Updates from Corma

Milestone$125M Series A at a $1B valuation led by NEA

Corma announced a $125 million Series A at a $1 billion valuation, led by NEA with Sequoia Capital, Khosla Ventures and Coatue participating, three months after it emerged from stealth with $60 million in funding. Total funding is $185 million. CEO Alon Pluda wrote that demand has been overwhelming, that securing compute capacity is a key challenge, and that the money will go toward scaling its foundation models and compute.

LaunchCorma emerges from stealth with a $60M seed round

Corma introduced itself in August 2026 as a lab training domain-specific foundation models for defensive cybersecurity, deployed as an AI security workforce, with a $60 million seed round led by Sequoia Capital. Alongside the launch it published research in which frontier models (Claude Opus 4.8, GPT-5.5, Grok 4.3 and DeepSeek V4), acting as attackers on a full-scale test enterprise network, planted a persistent backdoor in 85% of runs, while the same models acting as defenders found 19% of the implants.

Are you the founder? Claim this listing →