Scalable Capital Connects €60bn of Client Assets to Outside AI Assistants
Scalable Capital says it is the first European bank to connect accounts to outside AI assistants, and admits it does not yet know if it helps returns.
August 25, 2026 – 12:06 pm
Scalable Capital founders Erik Podzuweit (from left) and Florian Prucker.
Credit: Scalable Capital
Scalable Capital has opened its investment platform to ChatGPT and Claude, allowing clients to analyze their portfolios and place trades through the two AI assistants rather than through the broker’s own app.
The Munich company claims to be the first European bank to do this, and it arrives as the industry works out what handing trading capability to an AI agent actually looks like in practice.
The company is not small. Founded in 2014, Scalable has more than a million clients and over €60bn under management, operating chiefly in Germany and Austria with a presence in Italy, Spain, France, and the Netherlands.
The interesting decision is the sequencing. Most institutions build an assistant inside their own product first and expose it to third parties later, if at all, whereas Scalable has gone to the external platforms before integrating the capability into its own app.
Erik Podzuweit, the founder and co-chief executive, framed it as an opening move rather than a finished product.
“A lot of people might still be hesitant to let ChatGPT look at their portfolio, manage their portfolio,” he said, calling the launch “a first step.”
He was also unusually candid about the evidence base. His hypothesis is that using AI could produce better average returns, but he acknowledged that this remains to be demonstrated, which is a more honest position than most launches in this category adopt.
It’s important to note:
What has changed is that a chatbot that can read a portfolio is a convenience feature, while a chatbot that can execute a trade is a different category of thing. The distance between the two is measured in what happens when the model misunderstands an instruction.
Prompt injection is the failure mode that has no clean answer yet. An assistant that reads a portfolio also reads whatever text it is pointed at, and a model that can place orders is a considerably more attractive target than one that can only summarize.
Scalable says security measures protect customer accounts, without detailing them publicly. The obvious questions are how a trade instruction is confirmed, what limits apply, and who is liable when an assistant acts on an ambiguous request, and none has a published answer yet.
European banks have been moving towards conversational interfaces for a while, with bunq shipping its own generative AI assistant and Citi’s Jane Fraser arguing that two AI races will decide banking’s future. Almost all of that work has kept the model inside the institution’s perimeter.
Going outside it changes the regulatory picture. A German broker operates under MiFID II suitability and appropriateness rules, and it is not obvious how those obligations apply when the interface advising a retail client belongs to OpenAI or Anthropic rather than to the firm holding the assets.
Scalable presumably takes the view that the assistant is executing instructions rather than giving advice, which is a defensible line and also a fine one. Where a model summarizes a portfolio and a user acts on the summary, the boundary between information and recommendation is doing a lot of quiet work.
There is a distribution argument underneath all of it. Brokers spent the last decade fighting…