Cygrid Raises €2.66 Million to Control AI Agent Data Access

Cygrid, a Berlin cybersecurity startup founded in January 2026, left stealth on October 23 with a €2.66 million pre-Seed round to build AI agent access control that checks every action an identity takes before it executes.

The round was led by Atlantic, a pan-European pre-Seed fund. D11Z, the venture firm backed by the family office of the Schwarz Group, also took part. Angels include Boris Gorin, VP Product at Zscaler and founder of Canonic Security, Andreas Bermig, CEO of bonify, and Vlad Tropko of AmberVale Capital.

For security and compliance teams, the pitch targets a gap in how access is granted. Most identity and access management tools decide at sign-in and then stop watching. AI agents act on a user’s behalf for hours or days after that decision, often holding tokens that nobody reviews.

What Cygrid raised and who backed it

Cygrid will spend the money on expanding its enforcement capabilities, hiring, and building sales operations across Europe. The company said it aims to become the European alternative to US and Israeli providers of AI identity security.

Marc-Olivier Lücke, General Partner at Atlantic, said the founders combine “deep cybersecurity experience with a real understanding of what enterprises need: security that stops the wrong action without slowing the business down.”

Dmitry Kharchenko, chief executive and co-founder, and Dekel Paz founded the company. Both served in Israeli Military Intelligence.

The company reached 19 design partners across fintech, e-commerce and software-as-a-service in nine months. Six of them run Cygrid in production. In one early deployment, a Berlin-based direct-to-consumer company found AI agents and third-party app connections that its own security team did not know about. It used Cygrid to remediate more than 100 risky findings.

How the AI agent access control platform works

Cygrid enforces limits continuously instead of at login. It checks each action an agent takes and blocks or narrows it in real time, before the action runs.

The platform also maps human, non-human and AI agent identities into a single graph. For each identity, it correlates what the identity is entitled to, what it actually does, and on whose behalf it acts. Cygrid said this lets it revoke a forgotten token in seconds and stop a risky agent command before it executes.

“Access control tells you who can log in. It says nothing about what they do next,” Kharchenko said. “Once a token is valid, not one control in the stack can stop the action.”

That distinction matters for audit as well as for security. A control that records only who signed in cannot show what an agent did with the access it was given. A control that evaluates each action produces a record at the level where data is actually touched.

Why AI agent access control matters to risk teams

The number of identities that can act inside a company is growing faster than the people who review them. Palo Alto Networks’ 2026 Identity Security Landscape report found that machine identities now outnumber human ones by 109 to 1, and that 79 of those 109 are AI agents.

Microsoft’s Cyber Pulse report, published in February 2026, put adoption and oversight side by side. Microsoft said more than 80% of Fortune 500 companies use active AI agents built with low-code and no-code tools. The same report found that 29% of employees have used AI agents for work tasks without approval.

The attack data points the same way. A 2026 CrowdStrike report found that 82% of cyber detections involve no malware. Attackers instead used trusted identities, such as leaked tokens or compromised agent providers.

Gorin, Cygrid’s angel investor, works at Zscaler, a security vendor. His involvement does not change what the platform does, but it places the company inside an established network of identity and zero-trust buyers.

Where AI agent access control fits in the stack

Cygrid sells into a part of the security budget that is still being defined. Banks and insurers are already deploying AI agents for AML, fraud detection and risk monitoring. Each agent needs credentials, and each credential is a new path to confidential data.

Identity controls now sit alongside monitoring, logging and vendor tools in what compliance teams buy. A guide to what the modern compliance tech stack looks like shows where that layer sits relative to GRC and reporting systems. Boards have also begun to ask for it, as covered in what boardroom executives must prioritize on risk and cybersecurity.

Cygrid enters a field of young security vendors that have drawn capital this year. Nullify raised $12.5 million for product security, and Oneleet closed a $33 million round for security compliance. Neither competes directly with an identity enforcement layer, but all three pitch to the same security and compliance budget holders.

Established identity vendors are moving into the same area. Palo Alto Networks launched Idira on May 12, 2026, and it extends CyberArk technology to machine and agent identities. Cygrid’s position is that a European company built around agents from the start can move faster than those vendors.

What comes next for the Berlin startup

Cygrid has not named its design partners. Its first public customers will show whether buyers sit in the security organization, where the product is aimed, or in compliance and risk functions that need evidence of what agents did.

The company is hiring in Berlin and remotely across the EU, including go-to-market and security research roles. Its European sales push is the next milestone to watch, as the six production customers become reference accounts.