OpenAI’s 'Dots' Could become Banking’s Gatekeeper

OpenAI’s Dots could become a digital concierge through which customers coordinate decisions across banks and other providers. The deeper shift concerns who translates customer goals into financial instructions. Platforms could gain influence over provider selection while banks retain responsibility for accepting and executing the resulting actions.

A bank can hold a customer’s money and still lose its place in the customer’s decisions. If software compares providers, interprets goals and prepares instructions, the bank may encounter the decision only when it is asked to execute it.

OpenAI’s introduction of Dots on 29 September 2026 makes that prospect more tangible.¹ My thesis is that Dots could become the interface through which customers delegate economic authority across providers. OpenAI would not need to become a bank to influence which banks receive the business, while those institutions remain responsible for their execution decisions.

The Digital Concierge could define the Customer’s Choices

OpenAI describes Dots as an agent that continues assigned work between conversations, carries context forward and uses connected services within their permissions.² I would call this emerging role a Digital Concierge: a personal service that coordinates responsibilities across providers and acts within boundaries set by its user.

Imagine a Swiss household preparing to move. Its concierge could compare mortgages, review insurance, coordinate utilities and organise payments around one event. The bank would participate in a journey it no longer organises.

As of October 2026, this is a proposed financial use case. The OpenAI documentation reviewed here establishes neither a banking mandate standard nor autonomous Swiss account management.² My expectation that OpenAI will pursue this intermediary role is an inference from Dots design, not an announced company strategy.

In When Customers Ask AI First, I examined banks losing the default interface for financial decisions.³ In AI Agents as Financial Coaches, I explored continuous support and action within mandates.⁴ The further issue is who interprets those mandates when the interface operates across institutions.

The customer grants authority. The concierge translates objectives into actions. The bank decides whether it can accept the resulting instruction. Operating the interface does not make the platform the source of the customer’s legal authority.

But interpretation creates power. “Keep my money available” could mean immediate access or access within a week. That difference changes which products qualify and where funds go.

A future concierge might compare only accessible providers, favour commercial partners or receive referral fees. These are possible business models, not claims about OpenAI’s arrangements. The customer may delegate the work without noticing that the platform also defines the market they can see.

Know your Agent leaves a coordination Problem

Payment networks already address authority. In May 2026, Mastercard reported a live German agentic transaction, with identifiable agents, a verified customer mandate and transaction documentation among its stated controls.⁵ Visa’s Trusted Agent Protocol also describes signed agent recognition, consumer recognition and payment information.⁶ It would be inaccurate to portray these efforts as identity checks alone.

The harder problem is continuing delegation across providers. Suppose a concierge may transfer CHF 20'000 in total each month. Two banks each receive an instruction for CHF 15'000. Each falls below the headline limit, but executing both breaches the mandate.

Someone must maintain remaining capacity, reserve it for pending instructions and restore it when transactions fail. Withdrawing permission must also reach every institution still relying on it. Identifying the software and verifying a signed grant cannot resolve those coordination problems by themselves.

In We Digitised Life. Yet we left Death to the Heirs, I proposed Portable Trust for scoped estate authority.⁷ A concierge adds continuous interpretation and execution. Portable authority would need a current record of what remains permitted, alongside proof of the original grant.

Banks have good Reasons to refuse

The strongest objection is practical: a bank should reject an instruction when it cannot establish its scope or allocate responsibility for failure. A verified signature can establish evidence integrity. It cannot, by itself, establish that an action satisfies the receiving institution’s requirements.

“Optimise my finances” is also an inadequate mandate. Software could interpret acceptable risk differently from the customer, or satisfy a spending limit while undermining the wider objective.

Banks could restrict concierges to research and require direct confirmation for consequential actions. OpenAI’s documentation itself distinguishes connected access from permission to act.² That is a credible boundary, rather than an obstacle to progress.

Yet confirmation would not restore control over provider selection. The concierge could still gather offers, frame tradeoffs and present one recommendation. The bank would retain its right to refuse while another institution organised the choice.

The platform could control the shortlist, recommendation and timing. The bank could still be expected to detect an invalid instruction and explain its execution decision. This possible separation of influence and responsibility is the conflict banks should address before delegation becomes routine.

Banks should set the OpenAI Dots Terms before Platforms do

Authority.
Before connecting a concierge to execution, banks should define which delegated powers they recognise, which decisions require confirmation and who maintains shared limits. Banks and platform providers should agree how revocation reaches every participant and who bears losses when coordination fails. Permission inside a platform is insufficient evidence for execution by a bank.

Customer control.
Banks and concierge providers should make the translation from goals to permissions visible. Customers should be able to inspect and withdraw those powers, understand provider selection and see commercial incentives behind recommendations. Convenience should not require accepting an invisible shortlist or becoming dependent on one concierge.

Strategic choice.
Bank boards should establish their terms for accepting external mandates before integrations and customer habits make those terms difficult to change. They should assign responsibility for interpretation, selection and execution with participating platforms. Will banks help write the rules of delegated finance, or accept a future in which platforms direct the decisions and banks inherit the consequences?