Corma Raises $125M Series A for an AI Security Workforce, Unicorn in Three Months
Three months. That is how long Corma took to go from a $60M seed to a $1B valuation. On Monday the Tel Aviv and San Francisco company announced a $125M Series A led by NEA, with existing backers Sequoia, Khosla and Coatue back in. Total raised is now $185M. Primary coverage: https://calcalistech.com/ctechnews/article/bjiqapu0me and the company release https://www.businesswire.com/news/home/20261116743982/en/
Corma trains its own foundation model for defensive cybersecurity and deploys it as what it calls an autonomous AI workforce: a team of agents that sit on top of the security stack a company already runs, read logs, events and network traffic, run investigations and take preventive action. The founder and CEO is Alon Pluda, an alum of the Psagot program and Unit 8200. The team is about 25 people, mixing researchers from Google and DeepMind with Unit 8200 specialists.
The pitch is a sharp one. Models trained on code keep getting better at finding and exploiting bugs, which helps attackers first. Defense is a different job: thousands of sequential decisions over noisy telemetry, where one lapse in consistency is the breach. Corma bets that a general coding model will not get you there and a model built for the defender's long horizon will. It says customers include Fortune 100 and Global 500 companies in healthcare, finance, energy, retail and telecom, and that server capacity has grown 24x since August. No revenue figure was given.
The timing fits a week where the attacker's clock is visibly speeding up, with Google reporting the first fully automated patch-to-exploit pipeline in the wild. Investors are paying unicorn prices for the opposite bet. The open question is the one every security vendor faces: the claims here, including the goal of building superintelligence for defense, are the company's own, and nobody outside has scored the agents yet.
← Back to all articles
Corma trains its own foundation model for defensive cybersecurity and deploys it as what it calls an autonomous AI workforce: a team of agents that sit on top of the security stack a company already runs, read logs, events and network traffic, run investigations and take preventive action. The founder and CEO is Alon Pluda, an alum of the Psagot program and Unit 8200. The team is about 25 people, mixing researchers from Google and DeepMind with Unit 8200 specialists.
The pitch is a sharp one. Models trained on code keep getting better at finding and exploiting bugs, which helps attackers first. Defense is a different job: thousands of sequential decisions over noisy telemetry, where one lapse in consistency is the breach. Corma bets that a general coding model will not get you there and a model built for the defender's long horizon will. It says customers include Fortune 100 and Global 500 companies in healthcare, finance, energy, retail and telecom, and that server capacity has grown 24x since August. No revenue figure was given.
The timing fits a week where the attacker's clock is visibly speeding up, with Google reporting the first fully automated patch-to-exploit pipeline in the wild. Investors are paying unicorn prices for the opposite bet. The open question is the one every security vendor faces: the claims here, including the goal of building superintelligence for defense, are the company's own, and nobody outside has scored the agents yet.
Comments