Secure AI

Artificial Intelligence is becoming a cornerstone for the Swedish industry, the public sector, and society at large. But as adoption accelerates, so do the risks. Traditional cybersecurity focuses on protecting infrastructure, while AI Security is about ensuring that AI systems themselves behave safely, robustly, and ethically — even under manipulation or attack.

AI Sweden’s Secure AI initiative provides a unique national capability to:

Identify and respond to vulnerabilities in AI systems such as adversarial manipulation, data poisoning, and model leakage.
Develop and validate countermeasures that make AI trustworthy and resilient.
Help you understand how to keep the integrity of data you use to train AI, or you submit to an AI model.
Support compliance with the EU AI Act through concrete methods, tools, and evaluations.
Build talent and capacity by engaging researchers, students, and industry experts.

We have been working with AI Security since 2021, and Secure AI stands as Sweden’s collaborative hub for ensuring that AI systems remain safe, reliable, and aligned with our values — even in the face of evolving threats.

AI and Cyber Resilience

Sweden urgently needs to prepare for the imminent release of highly capable automated cyber attack agents. To meet and anticipate challenges, a group has been established as a cross-sector collaboration between government, industry, and academia.

Reports on Anthropic’s new ’Mythos Preview’ model should serve as a wake-up call for all decision-makers.

Robert A. (Bobby) Bridges

Innovation leader and AI security expert at AI Sweden

That was the message delivered to nearly 1,000 participants when Robert (Bobby) Bridges, joined John Billow, director of the Swedish National Cyber Security Centre (NCSC-SE), for a briefing on Mythos Preview and Project Glasswing on April 24, 2026.

Watch the recording:

More about AI and Cyber Resilience

Secure AI projects

Abstract blurry image, Gemini Nano Banana 2

The Swedish Center for Trusted and Resilient AI (STRAIC)

Artificial intelligence is already embedded in critical infrastructure like healthcare, energy, defense, and public services. But as AI becomes more pervasive, so do the risks: vulnerabilities in AI...

LeakPro 2: Operational privacy risk management for AI systems

LeakPro 2 is a tool/framework for assessing and mitigating privacy risks in machine learning models and AI systems. It combines privacy attacks, PET evaluation, and structured workflows to support...

Golden honeycomb pattern on a computer component with a metallic shine

AI-Powered Honeypots

Honeypots are decoy computing systems that mimic real environments to trick attackers into interacting with them. There are two ways a honeypot can increase security: (1) by occupying attacker...

AI Security Graduate Program

The AI Security Graduate Program is about talents securing robust, ethical, and sustainable AI, within your organization. Together with partners, AI Sweden and Nexer Tech Talent cultivate Sweden's...

Industrial Immersion Exchange program

AI Sweden, Dakota State University (DSU), and Chalmers University of Technology offer the opportunity to gain international and industrial experience in AI and cybersecurity as part of a travel and...