Inside Sygnum’s AI‑Agent Crypto Trading Pilot

The Human‑in‑the‑Loop Safeguard

Despite the automation, the AI agent does not have direct control of client assets.

Sygnum’s pilot kept a strict human‑in‑the‑loop model:

• The AI agent proposes and prepares transactions.
• Each step is evaluated for risk and shown to the client.
• The client must provide the final digital signature.

Without that signature, the transaction cannot be executed.

This design keeps the AI as a decision‑support and orchestration layer rather than an autonomous trading entity. It also preserves the accountability model expected in regulated banking environments.

Self‑Custody: Keys Stay With the Client

Another core feature of the system is self‑custody.

Transactions in the pilot were signed using wallets controlled directly on the client’s device. Private keys never left the user’s environment, ensuring the bank or AI agent could not independently move funds.

This architecture mirrors broader trends in crypto infrastructure where automation layers generate transaction proposals while the user retains final signing authority.

What Transactions the Agent Can Perform

The system focuses on multi‑step on‑chain financial operations rather than simple transfers.

Reports describing the pilot say the AI agent can plan and execute DeFi‑style workflows including:

• Stablecoin transfers
• Token swaps
• On‑chain lending transactions
• Liquidity provision in decentralized protocols

The agent determines the necessary steps automatically after interpreting the client’s objective.

For example, a user could describe an outcome—such as moving funds into a yield strategy—and the AI would determine the sequence of swaps, approvals, and protocol interactions needed to achieve it.

However, the full list of supported blockchains, assets, and DeFi protocols used in the pilot has not been publicly detailed.

Technology Behind the System

The pilot uses an AI‑agent architecture connected to blockchain transaction tools.

According to available reports, the system includes:

• An AI model that interprets natural‑language instructions
• An orchestration layer that plans transaction workflows
• Smart‑contract analysis and risk‑checking logic
• Wallet integrations for user‑signed transactions

Some coverage says the project used an internal Model Context Protocol (MCP) server within Sygnum’s AI initiative and ran on Anthropic’s Claude model to coordinate blockchain actions.

Through this architecture, the AI agent can access structured blockchain tools, analyze smart contracts, and prepare transaction sequences while keeping the human user responsible for final authorization.

Why This Matters for Banking

Sygnum has positioned the pilot as part of its broader AI@Sygnum strategy, which explores how AI agents can automate complex workflows across digital‑asset banking.

The experiment suggests a possible future where clients interact with financial infrastructure through natural language rather than specialized trading interfaces.

In theory, that could reduce the technical barriers that currently prevent many investors and institutions from using on‑chain financial products.

What Still Needs to Happen Before Client Rollout

The system remains a pilot rather than a generally available product.

Sygnum says commercialization will require further work in several areas:

• Regulatory and compliance approvals
• Security safeguards around AI‑generated transaction instructions
• Operational controls and client‑protection mechanisms
• Governance for automated financial decision workflows

The bank has indicated it plans to commercialize the service after addressing those regulatory and security considerations, but no public timeline has been announced.

The Bigger Trend: AI Moving Into the Transaction Layer

Sygnum’s experiment reflects a broader shift in financial technology. AI tools are moving beyond analysis and recommendations toward direct participation in financial execution.

If systems like this become widely deployed, interacting with crypto markets could eventually look less like operating trading software and more like describing financial goals in plain language.

For now, the Sygnum pilot shows how that future might work inside a regulated banking framework—where AI can orchestrate complex blockchain transactions, but humans still control the keys and the final decision.