Sequoia leads seed round for cybersecurity AI startup Corma

Sequoia Capital has announced its investment leading the seed round for Corma, a cybersecurity startup developing a specialized foundation model for enterprise defense. Founded by Alon Pluda, Corma aims to close a critical defensive gap created by general-purpose artificial intelligence. While offensive models like Anthropic's Mythos can identify and exploit zero-day vulnerabilities, defensive security has lagged. Standard large language models struggle with defense because security data, such as logs, telemetry, and traces, is rarely represented in their pretraining datasets.

To illustrate this disparity, Corma conducted red-and-blue team simulations where an attacker model planted a hidden backdoor. Even when using an identical copy of the attacker model as the defender, the defensive system failed to detect the backdoor 78 percent of the time. To overcome this, Corma is using large-scale reinforcement learning and self-play within simulated enterprise networks. This training pipeline produces defensive agents that outperform general-purpose models while offering significantly lower per-token inference costs and faster response speeds.

Corma has deployed its technology as an agentic Security Workforce across Fortune 500 companies and large enterprises in sectors like finance, healthcare, and critical infrastructure. These autonomous agents operate across various security tools to handle identity management, cloud security, and network operations. In one real-world deployment, a Corma agent detected and neutralized an active attack campaign within its first hour on the job, resolving a threat that the company's human security team had missed for 52 days. In another case, an agent alerted a chief information security officer of an ongoing attack via his Garmin watch, allowing him to authorize a shutdown with a single tap.

For cybersecurity practitioners, Corma's vertical integration and sovereign AI approach offer a shift from reactive monitoring to proactive, automated mitigation. By owning its model weights, the startup avoids the usage restrictions imposed by closed-source model labs. This specialization allows security teams to deploy always-on, highly fast defensive agents that can autonomously contain threats before they escalate, leveling the playing field against automated offensive AI.