Stopping a cyberattack while walking your dog
AI security startup Corma aims to close the defense gap in cybersecurity.
1Real-time defense example
Corma CEO Alon Pluda shares how a customer stopped a live attack while walking his dog, thanks to an AI agent notification. The security executive approved the agent's action, preventing the attacker from moving across the company network in under 10 minutes. The customer described this as 'one of the most magical moments of his year.'
2Company origin story
Pluda founded Corma about a year ago, naming it after the Elven word for 'ring' from Lord of the Rings. 'We're building the one ring to rule them all, but for defenders,' he says. Last week, the company announced $60 million in seed funding led by Sequoia Capital, with participation from Khosla Ventures and Coatue.
3Attack-defense imbalance
Corma tested four frontier models (Claude Opus 4.8, GPT-5.5, Grok 4.3, and DeepSeek V4) as both attackers and defenders. The models successfully implanted backdoors in 85% of attempts but detected only 19% of attacks. 'This shows the fundamental imbalance we're trying to solve,' Pluda explains.
4Defensive challenges
Pluda compares defensive security to 'finding needles in a haystack.' Unlike offensive tasks limited to code scanning, defense requires reading logs, events, configurations, and audit trails. Corma's research shows models read this structured machine data less reliably than prose or source code.
5AI workforce deployment
Corma's models power AI agents that organizations deploy like team members across security tasks. Fortune 100 and 500 companies in healthcare, finance, energy, and retail are using these agents. Early deployments reduced threat response times by over 94% and expanded security coverage 15x.
6Industry impact
Corma's solutions have uncovered multi-stage attack campaigns while significantly improving security metrics. 'With AI that's smart enough, domain-knowledgeable, and trustworthy, you can dramatically reduce these metrics,' says Pluda. 'You can cover far more ground than human intelligence alone allows.'
7Technological hurdles
Defensive reasoning is more open-ended than offensive tasks, which have clear goals like 'make this work' or 'break this.' Corma notes that models' training data and objectives naturally favor offensive security, creating the defense gap the startup aims to close.
8Future vision
Pluda states Corma's mission is to close the defense gap and ensure defenders win AI-powered security battles. 'Foundation models improve exponentially at offensive security but not at the same rate defensively,' he says, pointing to the industry's coming transformation in AI security capabilities.