Putting a Financial Product Inside Someone Else's AI Assistant: Authorisation, Branding and the Regulatory Perimeter
Most of the debate about AI agents and money is about access in the other direction: outside agents trying to get into your customers' accounts. The more consequential question for product leaders may be the mirror image. What happens when you put your own financial product inside an assistant that someone else controls? Our position is blunt. An assistant is a distribution channel, and you should treat it like one: with a contract, a control framework and a way to switch it off. If you cannot say who authorised a transaction, whose brand the customer thought they were dealing with and who answers to the regulator, you are not ready to ship.
The outbound mirror of agent gatekeeping
Two things are happening at once, and they are not as contradictory as they look.
On one side, incumbents are closing doors. JPMorgan Chase has reportedly blocked outside AI agents from direct access to customer accounts and backed open-banking routes instead, a move reported by Finextra citing CNBC. The same theme of platform gatekeeping has appeared with marketplaces. The debate has moved from whether agents can pay to who authorises a transaction and who is liable when it goes wrong, and that has been the dominant agentic-commerce theme for three consecutive reporting cycles.
On the other side, financial services and data providers are exposing products as agent-accessible tools inside assistants that other firms control. The example in the reporting is an investment-research provider making its tools available inside three different AI agents. That finding is a synthesis of a small number of reports and carries medium confidence, so we would not read it as a market-wide pattern yet. But it points to a direction, and the questions it raises are exactly the ones that gatekeeping raises: authorisation, branding and the regulatory perimeter.
The consistency is the point. A firm that blocks outside agents because it cannot tell who authorised an action has no business pushing its own products into other people's assistants without answers to the same question from the other side. Sanction your own routes before you ask for anyone else's.
The scale of distribution is why this is not hypothetical. Sensor Tower data reported by PYMNTS put Meta's Muse assistant above 10 million downloads in the United States and Canada within 35 days. Downloads are not active use, and the data covers North America only. But an assistant with a large installed base is exactly the kind of channel a product team will be asked to support.
Embedded finance has always meant someone else's interface
None of this is conceptually new. Embedded finance and digital wallets already put financial products inside other firms' apps, such as a retailer's checkout, a platform's wallet or a marketplace's payout flow. The partner owns the interface and the customer relationship; the financial firm owns the product and the regulatory responsibility. Product leaders have lived with that division for years.
What changes with an AI assistant is the nature of the interface. A retailer's checkout is deterministic: it renders what you designed. An assistant is generative. It decides what to say about your product, in what words, in what order, to which customer, based on a conversation you cannot see. The firm that built the product no longer controls how it is presented, and that is where authorisation, branding and perimeter all become harder.
Authorisation: keep the proof of consent on your side
The first rule is that the assistant's statement that a customer agreed is not evidence that the customer agreed.
When a customer asks an assistant to take out a product, move money or change a limit, your systems receive a request. They need to establish who is asking, whether the person has authority and whether this specific action was intended. An assistant that says "the user confirmed" is asserting something you cannot verify.
The reporting on agent identity gives a sense of how unsettled this layer is. In a single week in October, tooling for agent identity and delegated authority multiplied, including modules for agent authorisation and identity schemes such as Proof's x401. No single standard has emerged. For you, that means two practical things.
- Authenticate the customer on your own surface. For anything beyond a low-risk read, step the customer out to a journey you control, such as your app or an authenticated page, to confirm. The assistant can start the request. Your system should complete the authorisation.
- Record the mandate, not the conversation. Capture what was authorised, by whom, under what limits and at what time, in a form you own. A chat transcript held by someone else is a poor audit trail.
Consumer behaviour supports caution. Surveys reported in October found that willingness to use AI to research outpaces willingness to let it complete a purchase. Our recommendation follows from that. Launch with read-only and quote-style capabilities first, such as balances, eligibility indications or product explanations, and let delegated action arrive later, with limits, as trust develops.
Branding: the customer must know whose product this is
The second rule is that the customer should never be uncertain about who is responsible for the product.
When your product appears inside an assistant, three brands are in play: yours, the assistant's and perhaps the model provider's. In a conversation, the boundaries blur. A customer asking "is this safe?" may be answering to the assistant's voice while the commitment is yours.
Be explicit about three things.
- Attribution. Every response that describes or acts on your product should identify your firm as the provider, in language a customer would notice. Do not rely on an icon.
- Source of content. Product descriptions, rates, eligibility and terms should come from content you approved and published, not from whatever the assistant infers. If the assistant cannot quote you accurately, the right behaviour is to say so and link to your page.
- Handover. Decide what happens when a customer needs a person. Complaints, vulnerability and disputes need a route back to you, however they began.
There is a regulatory dimension too. On 19 October the FCA said that AI providers should add disclaimers to AI financial advice, published its view of good and poor practice in AI, and acknowledged agentic finance. As reported, it is applying existing frameworks, including Consumer Duty and senior-manager accountability, rather than writing new rules for agents. The disclaimer expectation is aimed at AI providers, but a financial firm cannot assume that someone else's disclaimer discharges its own duties to customers. That is a question for your compliance function, and we return to it below.
The regulatory perimeter: who is doing what, to whom?
This section is general context, not legal advice. The detail depends on your permissions, your product and the specific arrangement, and it should be reviewed by your compliance and legal teams before anything goes live.
The FCA's approach, as reported, is that existing frameworks apply. For a firm distributing through an assistant, that points to a handful of questions that the perimeter tends to turn on.
- What activity is taking place? Describing a product, arranging it, recommending it and executing it are different things, and they may carry different regulatory consequences. An assistant that moves from explaining to recommending may have changed the activity, and with it your position.
- Who is making the communication? If an assistant generates text about your product, whether that text is a communication you are responsible for is not something to assume either way. Work it through, with advice, before launch.
- Whose permissions are relied on? A firm that distributes through another party's assistant needs to be clear which entity holds the relevant permissions for each step, and that the arrangement is documented.
- Where does accountability sit internally? Under senior-manager accountability, someone senior needs to own the risk of an AI-mediated channel, in the same way they would own any other distribution relationship.
- How does Consumer Duty apply? If a customer receives poor outcomes from a journey that began in an assistant, the question of whether your product and communications supported good outcomes does not disappear because another firm ran the interface.
We are not suggesting an answer to any of these, because the answers differ by firm. We are suggesting that they are asked first, by the right people, in writing, before an engineering team is asked to integrate.
Six operating rules for a product inside someone else's assistant
Here is the framework we would put in front of a product and compliance leadership team. It is our own, and it is deliberately short.
- Expose capabilities, not accounts. Offer narrowly scoped tools, such as "get an indicative quote" or "check eligibility", rather than broad access to customer data.
- Authorise on your surface. Anything consequential is confirmed in a journey you control, with your authentication.
- Read before write. Start with informational capabilities and add actions only when you can evidence the controls.
- Serve approved content. Make the assistant's source of truth your published, approved material, with attribution to your firm.
- Log what you can prove. Keep your own record of every request, the mandate and the outcome, independent of the assistant provider's logs.
- Build a kill switch per channel. You should be able to disable one assistant without affecting others or your direct channels.
Behind those sits a contract. Agree with the assistant provider how your brand is presented, what data they may keep and use, what they do on an incident and who bears the loss when something goes wrong. Customer ownership belongs in that contract explicitly: if the customer first met your product in someone else's assistant, who holds the relationship afterwards?
What we do not know
This is an early market and the evidence is thin. The distribution finding rests on a small number of reports and carries medium confidence. We have not seen published data on conversion, complaints or losses for financial products sold through assistants, and we will not guess. The FCA's position is described in a speech and in reporting on it, and firms should read the primary sources, not summaries, before relying on any detail. Our recommendations are a point of view about how to be careful, not a forecast of how quickly this channel will grow.
Where Innovify fits
Innovify is a technology partner with deep expertise in banking, financial services, fintech and insurance. Putting products inside other platforms is the heart of embedded finance, and the assistant is the newest such platform. We design compliance in from day one, so the authorisation, audit and disclosure points described here are built into the product rather than added afterwards, and we build for regulated clients rather than being a regulated firm ourselves.
As an implementation partner across multiple fintech platforms, we help brands find a fast, compliant route to launching financial products. If you want to test an assistant-based journey before committing, Innovify AI Labs lets enterprise teams run several proofs of concept in parallel, test with real users and finish with a business case ready for funding.
FAQ
Can a bank or fintech distribute products through an AI assistant?
Yes, in principle, and some financial services and data providers are already exposing products as tools inside assistants that other firms control. What matters is how it is done. Authorisation, branding and regulatory responsibility need to be designed and documented before launch, and the arrangement should be reviewed by compliance and legal advisers.
Who is responsible when a financial product is delivered through another firm's assistant?
Responsibility depends on the product and the arrangement, but you should not assume that it transfers to the assistant provider. The FCA has indicated that it is applying existing frameworks, including Consumer Duty and senior-manager accountability, to AI. A firm that offers a regulated product remains accountable for its product and its customer outcomes, so agree the allocation of duties in writing.
Does a disclaimer from the AI provider cover the financial firm?
Not necessarily. The FCA has said that AI providers should add disclaimers to AI financial advice, but that expectation is aimed at the AI provider. Whether a disclaimer discharges any duty owed by the financial firm is a question for that firm's compliance function. Do not rely on it as a substitute for your own controls.
What should we launch first inside an AI assistant?
Start with read-only, informational capabilities such as product explanations, eligibility indications or quotes. Consumer surveys reported in October found that willingness to use AI to research outpaces willingness to let it complete a purchase. Add actions that move money only when you have your own authorisation, limits and audit trail in place.
How do we keep our brand clear inside someone else's assistant?
Require every response that describes or acts on your product to name your firm as the provider, and serve product facts from content you have approved. Define a handover to a person for complaints and vulnerable customers. Put the brand presentation, data use and incident handling into the contract with the assistant provider.
Conclusion
Financial products are starting to appear inside assistants that other firms control, and the questions that raises are the mirror image of the access debate: who authorised this, whose brand is this and whose responsibility is it? Treat the assistant as a distributor you do not control. Keep authorisation on your own surface, keep your brand and content attributable, start with read-only capabilities and put the regulatory perimeter questions to your compliance team before you build.
