Corma Raises $60M to Close AI 'Defensive Gap' in Cybersecurity
The startup is building a specialized foundation model to counter the growing advantage of AI-powered offensive attacks.
AI security startup Corma has secured $60 million in seed funding to build a foundation model dedicated to defensive cybersecurity. The company aims to neutralize a growing imbalance where general-purpose AI models are significantly more effective at launching attacks than mitigating them.
The funding round was led by Sequoia Capital, with participation from Coatue and Khosla Ventures. Based in San Francisco and Tel Aviv, Corma is led by CEO Alon Pluda and staffed by alumni from DeepMind, Google, and Israel's Unit 8200. The company's core product deploys AI agents as a "virtual workforce" capable of operating across an organization's existing security infrastructure to detect and block threats in real-time.
The Defensive Gap
The push for a specialized model stems from a critical disparity in how AI handles security tasks. According to Corma's research, frontier models—including Claude Opus 4.8 and GPT-5.5—demonstrated a stark bias toward offensive capabilities. In simulations, these models successfully implanted backdoors in 85% to 88% of cases, yet they detected attacks in only 12% to 19% of instances.
This "defensive gap" exists because general-purpose AI is highly proficient in coding and vulnerability research, which drive offensive security. Conversely, defensive security requires processing structured machine data, such as audit trails, logs, and on-disk state—information far less prevalent in the massive datasets used to train general models.
Industry Implications
As AI agents evolve from simple assistants into tools capable of executing end-to-end attack chains, the speed and scale of cyber threats are increasing exponentially. Corma argues that a specialized defensive model is necessary to match this sophistication. "We’re building the one ring to rule them all, but this time for the defenders to have this power," Pluda told The Register.
Early results suggest the approach is working. In deployments across Fortune 100 and 500 companies, Corma reports that its AI agents expanded security coverage by 15 times and reduced threat response times by more than 94%. Pluda describes the offering as "virtual human resources" capable of performing nearly any defensive cyber task.
The Path Forward
Corma's goal is to shift the balance of power back to security teams by providing an autonomous workforce that can operate at machine speed. While the initial seed funding provides a significant runway, the company must now prove that its specialized foundation model can maintain these performance gains as offensive AI continues to evolve. The industry will be watching to see if a dedicated defensive model can permanently close the gap or if the offensive advantage is an inherent trait of large-scale AI.