A European Bank Lets AI Assistants Handle Trades, With Your Approval

A European retail bank has opened a new route into its customers' investment accounts: an outside AI assistant can inspect a portfolio, prepare an order and manage investing tools through an ordinary prompt. Scalable Capital launched the feature on August 25, but the assistant cannot complete a trade until the customer approves it.

KEY FACT
Every securities order and savings plan requires the customer's explicit confirmation before execution.

From a prompt to a prepared order

Scalable says customers can use natural language to search stocks, exchange-traded funds and derivatives, examine portfolio risk, set price alerts, adjust limit orders and prepare savings plans. The bank exposes those functions through a Model Context Protocol server and a command-line tool, according to its launch announcement and security description.

The distinction between preparing and executing matters. Customers authenticate with Scalable credentials and two-factor authentication, then receive the same cost documents and transaction notices used by the bank's website and app. Deposits and withdrawals remain unavailable to the connected assistant.

Why this is more than chatbot convenience

A chatbot that summarizes a portfolio changes how information is displayed. A chatbot connected to brokerage tools changes what a mistaken or ambiguous instruction could trigger. Reuters confirmed that the service is live for a platform with more than one million customers and over €60 billion in client assets.

The bank calls the launch a European first, but that claim has not been independently audited. More important for customers, Scalable's chief executive told Reuters that any improvement in investment returns remains a hypothesis, not an observed result.

Where the risks begin

Scalable says outside AI applications operate independently and beyond its control. That division matters because a customer may be sharing sensitive portfolio data with an assistant whose outputs are not investment advice and whose mistakes remain the investor's responsibility.

The approval screen reduces the chance that a bad prompt becomes an unwanted trade, but it does not prove the preceding analysis is sound. The Next Web identified unresolved questions around ambiguous instructions, prompt injection and the boundary between information and advice. NewTqnia has previously explained how tool-using browser agents can weaken boundaries between services and how multi-step requests can bypass agent safeguards.

What happens next

The useful test will be operational, not promotional: how often users reject or correct prepared orders, how the system handles misleading outside content, and whether regulators treat assistant-generated analysis as information or advice. Finextra's launch coverage confirms that users can revoke access and monitor activity, but it does not answer those longer-term questions.

The takeaway

Scalable has turned an AI assistant into a new interface for a live brokerage account, while keeping final execution behind a human confirmation. The unresolved issue is whether that confirmation remains meaningful when the same assistant has selected, explained and prepared the order.