High-Integrity AI Control Planes for Swedish Datacenter Industry
| Reference number | |
| Coordinator | Kungliga Tekniska Högskolan - KTH Skolan för elektroteknik och datavetenskap, avdelningen för teoretisk datalogi |
| Funding from Vinnova | SEK 9 200 000 |
| Project duration | July 2026 - June 2029 |
| Status | Ongoing |
| Venture | Advanced digitalization - Industrial needs-driven innovation |
| Call | Shared, sustainable and secure AI infrastructure through advanced digitalization |
Purpose and goal
The project will make shared Swedish datacentres safer for AI. Its goal is to develop and test technology that lets operators and tenants verify the provenance, configuration and runtime behaviour of AI components. The result will be a reference integrity framework for the AI control plane, spanning formal modelling, cryptographic traceability, hardware-rooted attestation and enforceable controls.
Expected effects and result
The project will deliver an integrity framework and threat taxonomy, open-source prototype tools for verifiable artifact provenance, and a reference architecture for runtime integrity enforcement. The solutions will be validated at CanaryBit and Bahnhof. Guidelines, publications and knowledge transfer will strengthen Swedish datacentres and enable more efficient hardware sharing with lower energy use.
Planned approach and implementation
Over 36 months, KTH will coordinate research in five work packages. The project first models integrity requirements and threats, then develops prototypes for cryptographic binding of AI artifacts and a control-plane architecture with TEE-based attestation and runtime controls. CanaryBit contributes confidential-computing expertise and Bahnhof representative operating environments. The mechanisms are industrially validated and documented as open source and guidelines.