The Israeli-American start-up claims that general-purpose models succeed in 88 per cent of their simulated attacks but detect only 12 per cent of them when asked to defend against them.
Corma Labs, founded in 2025 and based in Tel Aviv and San Francisco, announced on 10 August 2026 that it had raised $60 million in seed funding led by Sequoia Capital, with Khosla Ventures and Coatue. The company describes itself as a cutting-edge AI laboratory working exclusively on the defensive side of IT security.
The figure underpinning this specialisation comes from the company’s own simulations, carried out on models of Fortune 500 corporate networks. When placed in the attacker’s position, the large language models from OpenAI and Anthropic successfully breach the systems in 88 per cent of cases. When tasked with defending the same environments, they intercept only 12 per cent of attacks. It is this asymmetry — a rapidly advancing offence, a defence that lags behind — that Corma says it aims to bridge with a foundation model designed for defence.
Its agents integrate with existing security tools rather than replacing them, and continue to learn about the environment in which they are deployed. The company reports that its early clients — in the healthcare, financial services, energy, critical infrastructure and retail sectors — have seen a reduction of more than 94 per cent in threat response times and a fifteen-fold increase in coverage. Its founding team brings together researchers with backgrounds at Google and DeepMind alongside specialists from Unit 8200 of the Israeli intelligence service.
Read the full article on the original source
Go to the source →