Corma CEO Alon Pluda said his AI security startup aims to close a defense gap where frontier models are stronger at offensive security than defensive tasks. He described a customer who, while walking his dog, received a notification on his watch that an AI agent had detected a live attack, sought permission to block it, and within 10 minutes stopped the malware and prevented further network movement. The customer called the experience one of the most magical moments of his year, Pluda recalled. Corma, founded a year ago and named after an Elvish word for ring, develops AI agents that operate as defensive security team members. The company announced a $60 million seed round led by Sequoia Capital, with participation from Khosla Ventures and Coatue. Corma works with Fortune 100 companies to train models for superintelligent defensive cybersecurity. In tests, four frontier models — Claude Opus 4.8, GPT-5.5, Grok 4.3, and DeepSeek V4 — were evaluated as both attackers and defenders in simulated enterprise networks. Across 241 engagements, the models implanted a persistent backdoor in 85 percent of attempts but detected attacks in only 19 percent. Pluda said the imbalance reflects faster improvement in offensive capabilities and limited ability of models to read structured logs and configuration data. He added that defensive security involves open‑ended reasoning and finding anomalies in large data sets, which current models handle less reliably. Early deployments have reduced threat response times by more than 94 percent, expanded security coverage by 15 times, and uncovered multi‑stage attack campaigns.