Corma raises $60M to build AI agents for cyber defense

Corma $60m funding was announced Aug. 10 as the Tel Aviv- and San Francisco-based startup emerged with a seed round led by Sequoia Capital. Khosla Ventures and Coatue also participated in the financing, which Corma says will support its effort to build a foundation model and AI agents for defensive cybersecurity.

The $60 million seed is a large opening round for a company founded in 2025, but financing does not establish the product’s effectiveness. The announcement materials supplied here do not identify customers, valuation, revenue or use of proceeds. Fortune, reporting on the round, said the company planned to expand the data and training behind its models and hire across defensive security, AI and research.

What does Corma say its cybersecurity AI agents do?

Corma says its agents operate across a customer’s existing security tools, analyze material including audit logs, security events and network flows, and learn an organization’s environment after deployment. Its argument is that general-purpose models’ strengths in coding and software reasoning translate readily to offensive work, while defense depends on connecting weak signals over time and making consistent decisions across many actions.

That is a product thesis, rather than independently established evidence of a gap between offensive and defensive AI. Corma says its team includes researchers from Google and DeepMind, along with cybersecurity specialists associated with Israel’s Unit 8200 and cybersecurity companies.

What evidence supports Corma’s performance claims?

Corma says it ran hundreds of simulations in enterprise environments modeled on Fortune 500 organizations, using leading AI models first as attackers and then as defenders. According to the company, attackers succeeded in 88% of those tests while defenders detected threats in 12%.

The simulation methodology and findings have not been independently verified, as Ynet reported. Corma also says its agents have been deployed for six weeks at Fortune 100 and Fortune 500 organizations across healthcare, financial services, energy, critical infrastructure and retail. It claims early deployments cut threat-response times by more than 94% and expanded coverage 15-fold across security functions. Those outcome figures are company-provided and have not been independently confirmed.

The supplied announcement does not provide the model architecture, training data, named reference customers, or an external benchmark for evaluating those results. For security teams, the near-term question is whether Corma can turn its claimed specialist training into repeatable outcomes inside production environments, rather than in its own simulations.

• Read Fortune’s reporting on the funding round
• Read Corma’s company-distributed ACCESS Newswire release

This story draws on original reporting from SiliconANGLE.