Sygnum became the first regulated Swiss bank to conduct live digital asset market transactions through an artificial intelligence agent while ensuring clients maintained complete control over custody, wallet authorization, and final transaction approvals. The pilot project represents a significant development in the integration of AI-driven workflows into regulated financial systems as banks increasingly evaluate how intelligent systems can interact directly with blockchain networks and decentralized finance infrastructure.
According to Sygnum, the AI agent was capable of interpreting plain-text instructions from clients and independently preparing complex blockchain transactions on a live Mainnet environment. Once the workflow was completed, the system presented the prepared actions to clients for final authorization before execution.
The pilot demonstrated how AI agents can autonomously prepare and orchestrate blockchain transactions while preserving full client ownership and control over digital assets.
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 explained that the architecture was intentionally designed around a human-in-the-loop framework, ensuring that AI-supported operations could not proceed without explicit client approval.
The distinction between supervised AI systems and fully autonomous AI agents has become increasingly important for regulators and banking supervisors. Financial authorities worldwide continue to examine concerns surrounding accountability, governance, operational resilience, cybersecurity risks, and the possibility of AI systems exceeding intended instructions.
Unlike traditional chatbots or analytics platforms, agentic AI systems are designed to independently execute multi-step tasks while dynamically adapting to changing instructions and environments. In Sygnum’s implementation, the AI agent was able to prepare 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.
.@finews_ch: Sygnum is the first bank regulated in Switzerland to launch tests with live transactions by an AI agent
In today's article by Stephan Waldvogel finews, Finews reports that this "reflects a core principle of the AI@Sygnum program, in which AI complements and… pic.twitter.com/AtEVTeYUv2
— Sygnum Bank (@sygnumofficial) May 18, 2026
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 Focus
Thomas Frei, Head of AI and Data Analytics at Sygnum Bank, reportedly stated that connecting AI agents directly to wallets represented a foundational step in the future evolution of finance. Frei indicated that financial markets could eventually see AI agents transacting, settling, and interacting with markets continuously on behalf of clients over the coming decade.
He further explained that the institution’s primary objective involved balancing automation with bank-grade consent, trust, and custody protections. Frei noted that the pilot was designed to demonstrate how a regulated bank could provide faster and more accessible AI-driven execution services without compromising client control over assets.
The architectural question for AI in banking has always been the same: how do you give an agent meaningful capability without compromising self-custody?
This week's @Finextra coverage is our answer. The pilot Sygnum has just run uses an AI agent, powered by @AnthropicAI's… pic.twitter.com/OBJR6F9IPQ
— Sygnum Bank (@sygnumofficial) May 18, 2026
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.
Open Standards and AI Infrastructure Expansion
The pilot was developed using a Model Context Protocol server created internally by the Sygnum team and powered by Anthropic’s Claude model. Model Context Protocol, commonly known as 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.
The initiative also highlighted how blockchain networks are increasingly becoming testing environments for agentic AI systems because programmable assets and smart contracts naturally support automated financial workflows.
The broader financial industry continues to debate how much operational authority AI agents should ultimately receive within trading, treasury management, settlement, and portfolio management systems. While AI promises faster execution and continuous market interaction, regulators remain cautious about unintended actions, governance failures, and uncontrolled automation. Sygnum’s pilot positions regulated banking infrastructure between traditional finance and fully autonomous AI-native markets by combining AI-driven orchestration with mandatory human oversight.
