Corma CEO says AI defenders can stop attacks in minutes

Corma, a cybersecurity startup founded about a year ago by Alon Pluda, is developing AI agents designed to perform defensive security tasks. Pluda told The Register that while large language models from OpenAI, Anthropic, and Google are exceptionally good at offensive security—such as finding and fixing bugs, orchestrating tools, and even conducting end-to-end attacks—they are significantly less capable at defensive tasks like reading logs, monitoring events, and detecting intrusions. He cited a benchmark where models successfully planted persistent backdoors in 85% of runs but detected only 19% of attacks, highlighting what Corma calls the "defensive gap." To address this, Corma trains its own models specifically for defensive security, enabling AI agents that act like team members. These agents can be assigned to various security tasks, and in one customer story, a security executive received a notification on his watch while walking his dog, alerting him to a live attack. The agent requested permission to block it, and upon approval, blocked the malware and stopped the attacker's lateral movement in under 10 minutes. Pluda described this as "one of the most magical moments of his year" for the customer. Corma claims to work with Fortune 100 and 500 companies across healthcare, financial services, energy, critical infrastructure, and retail. Early deployments reportedly reduced threat response times by over 94%, expanded security coverage by 15 times, and uncovered multi-stage attack campaigns. Pluda emphasized that with sufficiently intelligent and trustworthy AI, organizations can significantly improve their security posture and cover more ground than with human intelligence alone. The startup's mission is to ensure defenders win in the intelligence-versus-intelligence game of cybersecurity.