Sygnum became the first regulated Swiss bank to conduct live digital asset market transactions through an artificial intelligence agent while maintaining client control over custody, wallet authorization, and final transaction approvals. The pilot project, announced May 18, 2026, represents a significant development in integrating AI-driven workflows into regulated financial systems as banks evaluate how intelligent systems can interact directly with blockchain networks. According to Sygnum, the AI agent interpreted plain-text instructions from clients and independently prepared complex blockchain transactions on a live Mainnet environment, then presented prepared actions to clients for final authorization before execution.
Human-Controlled AI Execution Model
Sygnum emphasized that private keys never left client possession during the process. All transactions were signed exclusively through self-custodial wallets operated directly by clients on their personal devices. The bank designed the architecture around a human-in-the-loop framework, ensuring that AI-supported operations could not proceed without explicit client approval.
The AI agent was capable of preparing workflows involving stablecoin transfers, token swaps, liquidity provisioning, token wrapping, and decentralized lending positions. The system also reviewed smart contracts, assessed transaction structures, evaluated potential risks, and mapped execution strategies before requesting client authorization.
Sygnum structured the platform so that AI handled workflow preparation and operational analysis while clients retained exclusive authority over transaction approval and asset custody.
Governance and Regulatory Framework
Thomas Frei, Head of AI and Data Analytics at Sygnum Bank, stated that connecting AI agents directly to wallets represented a foundational step in the future evolution of finance. Frei indicated that the institution's primary objective involved balancing automation with bank-grade consent, trust, and custody protections, demonstrating how a regulated bank could provide faster and more accessible AI-driven execution services without compromising client control over assets.
Sygnum stated that its governance framework incorporated strict standards for transparency, accountability, data management, operational resilience, and risk oversight in line with regulated banking requirements. The bank stressed that AI was intended to enhance human decision-making rather than replace client supervision or trusted financial relationships.
Technology Infrastructure
The pilot was developed using a Model Context Protocol (MCP) server created internally by the Sygnum team and powered by Anthropic's Claude model. MCP is an emerging open standard intended to improve how AI systems and financial platforms exchange structured operational information.
The infrastructure enabled the AI agent to interpret wallet states, smart contract interactions, transaction environments, and workflow requirements while maintaining strict client-controlled safeguards. Sygnum described the system as both model-agnostic and asset-class agnostic, suggesting that it may eventually support multiple AI systems and a broad range of financial products.
