TrustAgent: Trustworthy Agentic AI for Secure, Transparent and Resilient SME Financial Operations
| Reference number | |
| Coordinator | Orchestable AB - Orchestable |
| Funding from Vinnova | SEK 1 690 000 |
| Project duration | June 2026 - December 2027 |
| Status | Ongoing |
| Venture | Advanced digitalization - Industrial needs-driven innovation |
| Call | Digital Resilience, AI and Cybersecurity |
Purpose and goal
The purpose is to make AI agents that handle finance and accounting in small and medium-sized enterprises secure, transparent and auditable. The project develops a trust layer that verifies every AI output before it is passed on, a decision log that explains each decision in line with the EU AI Act, and governance with human oversight. The goal is to stop errors and manipulated data from spreading between cooperating agents and to reach TRL 6 in real operation.
Expected effects and result
Results include a security protocol, an open attack-simulation dataset, a decision-explanation system with energy measurement per AI decision, automated compliance reporting, a governance framework, a bias detection tool and a peer-reviewed article. Tooling and dataset will be released openly. The expected effect is that Swedish SMEs can use agentic AI in financial processes with lower risk, better traceability and easier compliance with the EU AI Act.
Planned approach and implementation
The project runs from June 2026 to December 2027 in four work packages: threat modelling and verifiability, explainability and the EU AI Act, validation including a four-month live pilot, and dissemination with open release. Orchestable coordinates and develops the platform, RISE performs independent validation, Södertörn University contributes to pilot design, market analysis and dissemination, and an accounting firm and SME partners test the solution on real data.