DIREC project

Security by Design for AI Startups: Secure and Scalable AI Agents

Project impact

The project aims to identify tools and frameworks suited for startups to build security into AI agents right from the start.

The scientific approach is based on a case study in which the Alexandra Institute uses PrivacyMate to systematically map security challenges faced by AI-based startups.

At the same time, the Alexandra Institute will use its accumulated specialist knowledge to help enhance the security of other Danish AI startups.

PROJECT DATA

Project name

Security by Design for AI Startups

Project period

2025-2026

Funding

DKK 400.000

Scientific mission

The project aims to embed security into AI agents right from the start. The improvements will initially be integrated into PrivacyMate’s AI agent Hipako, but can also be applied to other products.

The scientific approach is based on a case study in which the Alexandra Institute uses Hipako to systematically map the specific security challenges faced by AI-based startups.

The results will be directly implemented in PrivacyMate’s operations, strengthening its core product. The company will be able to help even more customers streamline their compliance with data protection regulations and GRC (Governance, Risk, and Compliance) processes.

At the same time, the Alexandra Institute will use its accumulated specialist knowledge to help enhance the security of other Danish AI startups.

Project Participants

	Sebastian Christophersen – Strategic Programme Manager – Alexandra Institute
	Alberte Krogh – Techno-Anthropologist – Alexandra Institute
	Benjamin Salling Hvass – Security Architect – Alexandra Institute
	Zaruhi Aslanyan – Senior Security Architect – Alexandra Institute
	Panagiotis Vasilikos – Founder – PrivacyMate

Activities

WEBINAR

Security and testing tools for AI developers

This webinar helps AI developers avoid common security pitfalls in AI applications and shows how automated testing can be used to identify and mitigate vulnerabilities.

WEBINAR

Navigating the most common threats in AI-based systems

This webinar covers common threats to AI systems, from prompt-level attacks to adversarial inputs, and shows how organisations assess AI risks using threat modelling. It highlights the MAESTRO framework for analysing risks in Agentic AI systems. An AI-focused startup shares practical insights from applying MAESTRO in real-world systems, including lessons learned and challenges faced.