Corma привлекла $60 млн от Sequoia: GPT и Claude проигрывают защиту в 88% атак

Startup Corma raised $60 million in seed investment from venture fund Sequoia in August 2026 to develop defensive AI — a new layer of corporate cybersecurity that, according to the company, is virtually absent from the market.

What hundreds of attack-and-defense simulations revealed

The Corma team conducted hundreds of simulations built on profiles of major Fortune 500 corporations — with dozens of real security tools in each environment. In every simulation, leading language models were tested, including GPT (OpenAI) and Claude (Anthropic): first, those same models were tasked with attacking the organization and planting persistent hidden threats, then they were asked to switch roles — to find what they themselves had planted.

The result was telling: the attackers won in 88% of cases.

• Seed round: $60 million, investor — Sequoia
• GPT (OpenAI) and Claude (Anthropic) were tested alternately as attacker and defender
• Environment: hundreds of simulations based on Fortune 500 profiles with dozens of security tools
• Result: the attacking AI won in 88% of simulations

This result does not speak to the weakness of specific models — GPT and Claude performed excellently as attackers. It points to a structural asymmetry: offensive AI in cybersecurity is evolving rapidly, while the defensive side remains in its infancy.

Why defensive AI lags behind offensive AI

Most modern corporate cybersecurity solutions operate on a "detect and respond" model: they analyze incoming traffic, look for signatures of known threats, and raise alerts. A proactive defensive agent — one that independently explores infrastructure, forms hypotheses about attack vectors, and autonomously closes vulnerabilities — is virtually absent from the market.

Corma positions itself precisely in this niche. According to the company's reasoning, SOC (Security Operations Center) teams can no longer handle the volume and speed of threats manually, and the only scalable response is an AI agent that acts like an experienced analyst — without fatigue and with full coverage.

According to

Corma data published by The Next Web, the attacking AI wins in 88% of simulated scenarios — including cases where the model itself planted the threat and then searched for it.

Who competes with Corma

The defensive AI segment for corporate security has yet to be claimed by a single dominant player, but pressure is mounting. Microsoft promotes Security Copilot on top of its cloud ecosystem; CrowdStrike and Palo Alto Networks are embedding AI layers into SIEM and SOAR platforms. Corma's advantage is its clean focus on proactive defense and $60 million in resources from the outset, backed by one of Silicon Valley's most authoritative funds.

A seed-stage round from Sequoia of this size is rare: typically, such sums are raised at Series A. This signals that the fund is evaluating not just the product, but the size of the unoccupied market — corporate cybersecurity globally generates hundreds of billions of dollars annually.

What this means

Corma's simulations are not only a marketing case study, but an argument for the entire industry: offensive AI has already become an industrial tool, while the defensive layer is critically lagging. If the company succeeds in closing this gap, it will secure a position in one of the fastest-growing corporate AI segments.