Defensive AI Startup Targets the Offense Gap

Alon Pluda, CEO of AI security startup Corma, recalls a customer, a security executive, who was walking his dog when his smartwatch buzzed with a notification from a Corma agent. The message read, 'It said, 'I just caught a live attack. I need your permission to block it,'' Pluda told The Register in an interview. The executive approved the action, and the agent blocked the malware, halted the attacker's lateral movement, and mitigated the intrusion in under 10 minutes.

Pluda says the customer later described the experience—'walking outside with his dog, and blocking a real-live attack with his AI coworker'—as 'one of the most magical moments of his year.' That anecdote illustrates what Corma aims to achieve: closing what it calls the 'defensive gap,' where AI models are significantly better at offensive security than defensive security.

One Ring to Rule Them All

Corma, founded about a year ago, takes its name from the Elven word for 'ring.' Pluda jokes, 'We're building the one ring to rule them all, but this time for the defenders to have this power.' The company announced $60 million in seed funding this week, led by Sequoia Capital, with participation from Khosla Ventures and Coatue. It is already working with Fortune 100 companies, training models to achieve 'superintelligence for defensive cybersecurity.'

Pluda explains that frontier models from OpenAI, Anthropic, and Google are 'amazingly good' at coding and language, including finding and fixing bugs and orchestrating multi-step workflows. 'When you combine it with agentic capabilities, they move from being incredible vulnerability researchers to end-to-end attackers,' he said. 'So inherently, what we've seen in the last few months is the models getting exponentially better at offensive security, like we saw with the OpenAI and Hugging Face incident.'

The Defensive Gap Defined

But these same models are not as skilled at defensive tasks that go beyond code scanning. 'The vast majority of defensive security tasks don't have anything to do with code,' Pluda notes. Defensive work involves reading logs, events, configurations, audit trails, and on-disk state—'structured machine data that is neither prose nor source, and a small share of what these models see in training,' according to Corma's research. 'They appear to read it less reliably.'

Furthermore, defensive reasoning is open-ended, while offensive tasks have straightforward goals—like 'make this work' or 'break this'—and a checkable finish. This imbalance is what Corma aims to correct.

Testing the Frontier Models

Corma recently tested four frontier models—Claude Opus 4.8, GPT-5.5, Grok 4.3, and DeepSeek V4—as both attackers and defenders on a simulated enterprise network built to mirror a multi-business company. The attacker's task was to plant a backdoor; the defender's was to find and stop it. Corma ran all four models against each other in every pairing, including each model against itself, with 15 independent engagements per pairing for 241 scored engagements.

The results highlighted the gap: across all engagements, models successfully implanted a persistent backdoor in 85 percent of their runs, but only detected 19 percent of attacks. 'That speaks to the inherent imbalance we are trying to solve,' Pluda said. 'The general foundation models are getting exponentially better at offensive security, but haven't been able to improve on the same rate on defensive security. So our mission is to close this gap, and make sure the defenders win in this intelligence-versus-intelligence game - or war.'

Agentic Defenders

Corma's models power AI agents that organizations deploy 'like team members' to operate across defensive security tasks. 'It's a generalized workforce, and you can assign it to whatever security tasks you want,' Pluda says. Fortune 100 and 500 organizations across healthcare, financial services, energy, critical infrastructure, retail, and other sectors have deployed Corma's AI workforce, according to the startup.

The company reports that these early deployments have reduced threat response times by more than 94 percent, expanded security coverage by 15 times across different security functions, and uncovered multi-stage attack campaigns.

Why It Matters for Defenders

The implications for enterprises are significant. As AI models become more capable of offensive operations, the window for human defenders to react shrinks. Corma's approach suggests that AI can level the playing field, but the current gap means organizations relying solely on traditional defenses may be exposed.

Pluda's pitch is straightforward: 'If you can get AI that is smart enough, intelligent enough, knows the domain enough, optimizes for the right things enough, and you can actually trust it, end to end, all the way to responding to real-live attacks, you can reduce all of these metrics significantly,' he said. 'And you can cover way more ground than what is possible with just human intelligence.'

For security teams, the takeaway is that AI is becoming a double-edged sword: attackers are leveraging it, and defenders must adopt similar technology to keep pace. The dog-walking incident is a glimpse of what defense could look like—automated, responsive, and always on.