Cygrid raises €2.66M to prevent security breaches caused by stolen identities.

Key facts

• €2.66 million pre-seed led by Atlantic, with d11z and angels from bonify/SCHUFA, Zscaler and AmberVale Capital

• One platform for every identity: visibility, intelligence and enforcement for humans, non-human identities and AI agents

• Already in production at companies such as Barfer’s and Horizn Studios

• Built by cybersecurity and intelligence veterans in Berlin. Identity, enforced. AI, unlocked.

BERLIN – 22 October 2026 – Cygrid had already stopped more than 2,000 risky identity actions at over 20 companies, even before going out of stealth. Most cyberattacks start with a stolen identity. Now enterprises are deploying AI agents with deep system access — and every new agent is another identity to compromise. Think of an AI agent reaching for data it was never meant to see, or an app still holding admin access long after anyone needed it. Each of these came from an identity that was allowed to log in, so a login check alone would have let it through.

It’s the same pattern behind this year’s biggest breaches. At Vercel, it took one employee and one AI app. When the app was hacked in April 2026, attackers didn’t need a password: they simply used the app’s own access to get into Vercel’s systems. Two months later, a single forgotten credential at Klue opened the door to its customers’ Salesforce data. The front door was locked. The attackers came in wearing a trusted badge. Most organisations can’t list their non-human identities, cannot say what each one can reach, and have no way to stop it.

Cygrid is the platform that secures every agent, at every point where it can act. It maps every identity in a company, from employees and service accounts to AI agents, decides based on Cygrid Intelligence what each one should be allowed to do, and controls every executed action on sensitive information. A forgotten token is revoked in seconds. An AI agent’s risky command is rewritten in milliseconds or blocked, before it ever runs.

Cygrid, an AI-native identity enforcement platform, has today emerged from stealth with €2.66 million pre-seed led by Atlantic, with participation from d11z, as well as well-known angels including Andreas Bermig (CEO of bonify/SCHUFA), Boris Gorin (VP Product at Zscaler), Vlad Tropko (AmberVale Capital) and other enterprise leaders. The funding is being used to expand the platform’s enforcement capabilities, grow the team and scale go-to-market across Europe and beyond.

Your AI apps and agents have the keys. The question is who else does.

CrowdStrike’s 2026 Global Threat Report shows that 82% of the attacks it detected used no malware at all. Attackers simply borrow an identity that is already trusted, and the fastest moved from first access to other systems in just 27 seconds. This year reads like a list of proof:

• Klue, June 2026. An old credential nobody was watching unlocked the tokens Klue held for its customers’ Salesforce and Gong accounts. Attackers then queried customer CRM data at scale.

• Vercel, April 2026. One employee connected one AI app to their work Google account. The app got hacked, and its token became a skeleton key to internal systems and secrets. The attackers then asked for $2 million.

• PocketOS, April 2026. An AI coding agent hit a credential problem and decided to fix it on its own. In 9 seconds, with a single API call, it deleted the car-rental software company’s production database and all its backups. “I didn’t verify,” the agent admitted afterwards. The company had to restore from a three-month-old backup.

• Meta, March 2026. No outside attacker at all. An internal AI agent answered an employee’s question without permission, and following its advice exposed massive amounts of company and user data to engineers who had no right to see it. It took two hours to close, and Meta rated it “Sev 1”, its second-highest severity level.

These incidents show why companies need a new layer of control: one that understands what every identity, human or AI, is allowed to do, and steps in when it goes further. That is what Cygrid provides, so organisations can adopt AI safely, without slowing down.

A billion new AI employees, and nobody checking what they do

“The IT department of every company is going to be the HR department of AI agents in the future,” Nvidia CEO Jensen Huang predicted at CES 2025. That future has arrived faster than most security teams can hire. More than 80% of Fortune 500 companies are already deploying AI agents, and 29% of employees have used agents their company never approved, according to Microsoft’s Cyber Pulse report (March 2026). IDC expects 1.3 billion AI agents at work by 2028.

The people guarding the world’s most sensitive systems are sounding the alarm. Token-based integrations create “direct, often unchecked interactions between third-party services and firms’ sensitive internal resources,” warned JPMorgan Chase CISO Patrick Opet in an open letter to the bank’s suppliers. Microsoft’s security chief Charlie Bell goes further: “Containment simply means we do not blindly trust our AI Agents, and we significantly box every aspect of what they do.”

The numbers back them up. Machine identities, including AI agents, already outnumber humans 109 to 1 (Palo Alto Networks, 2026), and securing them is now the number one worry for 78% of CISOs (Team8, September 2026). Mid-sized companies face the same threat as the Fortune 500, with a fraction of the security team.

That’s the gap Cygrid fills. Companies don’t need to slow down on AI. They need a way to say yes to agents and still stop the one action that shouldn’t happen. Cygrid connects in minutes, builds your Identity Intelligence Graph in hours, and enforces from day one.

Every AI agent is an identity. Cygrid enforces its limits.

Cygrid connects to the existing stack, read-only first, and maps every identity, employees, service accounts, API connections and AI agents, into one Identity Intelligence Graph. The graph scores risk continuously and decides what each identity may do. Cygrid then enforces that decision wherever an identity can act: before a credential is issued, after it exists and at runtime, from revoking a stale grant in seconds to rewriting or blocking a tool call in milliseconds. One graph decides. Enforcement stops the action.

“Access control tells you who can log in. It says nothing about what they do next. Every identity in an enterprise can act, and once a token is valid, not one control in the stack can stop the action. Cygrid is the platform that does: it decides what each identity may do and enforces it before anything executes. This funding allows us to scale that approach and help organisations move from reactive identity security to continuous, real-time enforcement, so they can adopt AI agents at full speed without losing control of what those agents can do.” – Dmitry Kharchenko, CEO and Founder, Cygrid

Since its founding in January 2026, Cygrid has more than 20 design partners across fintech, e-commerce and SaaS, including Barfer’s and Horizn Studios, and runs in production on Google Workspace, Microsoft 365, GitHub, Slack, HubSpot and other estates. Cygrid won Germany’s federal AI Nation programme grant and was selected for PwC Germany’s Responsible AI programme.

“Cygrid showed us what each identity could actually do, and now stops what shouldn’t happen. We are adopting AI safely, in practice,” said Olga Khorkova, CTO and Board Member at Barfer’s.

Built by cyber operators who ran identity at scale

Founders Dmitry Kharchenko (CEO) and Dekel Paz (Chief Security Architect) both served in military intelligence. Kharchenko was CTO at bonify (acquired by SCHUFA), CTO/CDO at Friendsurance (acquired by Cover Genius) and Tech Leader at Firelayers (acquired by Proofpoint). Paz brought his intelligence background into security research, leading teams at Zero Networks, Novartis and LightCyber (Palo Alto Networks), and building open-source tools used by security practitioners worldwide.

Supported by top VCs and industry leaders

“Cygrid is solving one of the most critical challenges in modern cybersecurity: not just seeing identity risk, but stopping it. Where existing tools report, Cygrid enforces, whether the identity behind the action is an AI agent, a machine or a human. CISOs and CTOs can say yes to AI agents and still stop the one action that shouldn’t happen, on a platform that runs in hours, not quarters. That positions Cygrid to become the foundational enforcement platform for identities in the AI era.” – Boris Gorin, VP Product, Zscaler, and founder & CEO, Canonic Security

“Dmitry and Dekel are proven builders who combine deep cybersecurity experience with a real understanding of what enterprises need: security that stops the wrong action without slowing the business down. AI agents are widening the identity attack surface faster than any control can keep up. Intelligence-driven enforcement is the only answer that scales. The timing for Cygrid couldn’t be better.” – Marc-Olivier Lücke, General Partner, Atlantic

Enforcing identity where AI acts, from one graph

Cygrid builds a single intelligence graph across human, non-human and AI agent identities. The graph correlates what each identity is entitled to, what it actually does, and on whose behalf it acts. Access is modelled at the level of SaaS actions rather than roles, so risk can be compared across tools. The graph combines deterministic intelligence (policy and rules) with non-deterministic intelligence (AI-driven behavioural and contextual analysis). The graph decides what is allowed, and Cygrid enforces that decision in three modes:

• Pre-use: before an agent, app or integration receives access, the request is evaluated against the graph and policy, taking in its context, behaviour and who it acts for, rather than approved from static assignments.

• Out-of-band: policy-driven, automated enforcement across connected identity and SaaS systems. When an identity, permission or configuration drifts from policy, Cygrid corrects it at the source, without waiting for a review cycle.

• In-band: the Cygrid MCP Gateway evaluates every AI agent tool call against the graph and policy before it reaches the target system.

What makes Cygrid different is that it combines deterministic and AI-driven intelligence in one graph and uses that graph to drive pre-use, out-of-band and in-band enforcement together.

About Cygrid

Cygrid is the identity enforcement platform for AI-driven enterprises. It maps human, non-human and AI agent identities into one graph, scores what each one can do, and enforces it wherever an identity can act. Founded in January 2026 and headquartered in Berlin, Cygrid GmbH is backed by Atlantic Labs, D11Z.Ventures and leading cybersecurity operators. Identity, enforced. AI, unlocked. www.cygrid.io

Media contact

Masha Kovaleva
PR & Marketing Manager at Cygrid
masha.kovaleva@cygrid.io