Sygnum has tested an AI agent for live digital asset transactions, allowing clients to execute complex blockchain trades through plain text commands while retaining control of their private keys.

The Swiss regulated bank used the tool to plan actions on the blockchain, review smart contracts, and flag potential risks before asking the client for final approval.

The transaction is only signed through a self custodial wallet on the user’s device.

The system can handle stablecoin transfers, asset swaps, lending positions, token wrapping, and liquidity provisioning.

Sygnum built the pilot using Anthropic’s Claude and a Model Context Protocol server, an open standard enabling data sharing between AI systems and financial platforms.

“Connecting AI agents to wallets is foundational to where finance is heading,”

said Thomas Frei, Head of AI and Data Analytics at Sygnum Bank.

He noted that the challenge for the coming decade is enabling automated execution without compromising custody and trust.

Unlike setups where AI agents transact autonomously from their own wallets, the Sygnum architecture requires direct human involvement.

The bank highlighted that regulators increasingly view autonomous AI agents as a risk for financial institutions.

Sygnum designed the system to meet strict regulatory standards by keeping consent and custody with the client.

The pilot aligns with the broader AI strategy at Sygnum, which focuses on augmenting client experiences and improving operational efficiency rather than replacing human decisions.

The bank confirmed the agent is currently in a controlled testing phase and not yet available to clients.

Featured image credit: Edited by Fintech News Switzerland, based on image by tahantanha10 via Magnific