Your Money App Should Not Be Trying to Keep You

The Consumer Finance AI Standard carries no legal force. Consumer Reports is an advocacy organization, and the document it published on June 30 is its own proposal for what people should be able to expect when AI runs inside their financial lives. No regulator has adopted a word of it. I still think it is the most useful thing written about AI and money this year, because one of its nine principles says out loud something the industry has been careful to leave unsaid.

The principle is called Duty of Loyalty, and the announcement states it this way: "AI financial products must be designed and operated to advance the financial interests of the consumer as their primary objective, rather than for the entity, its commercial partners, or third parties." The short form on their explainer runs eight words: "The AI works for you, not the company."

I circled a nearby question back in June, in can you trust AI for financial advice. That post was about the line between advice and education, and about whether an AI carries the fiduciary duty a licensed adviser does. This one is about the question sitting underneath that: whose interest the system was built to serve. An assistant can be right about every number it reports and still be aimed at the wrong outcome.

Agreement is the cheapest way to keep you talking

A person bound by a duty of loyalty is obliged to talk you out of a bad decision. Telling someone that the plan they walked in excited about will hurt them is the least pleasant part of that job, and it is most of what the duty is for.

An assistant tuned to keep a conversation going has the reverse incentive. Agreement is smooth, it costs nothing, and it produces another message. Disagreement risks ending the session. Engagement is easy to measure, and a number that easy to measure shapes what gets built long before anybody argues about it in a meeting. Nobody files a ticket asking the model to flatter the user. The behavior shows up anyway, because the thing being measured rewards it.

If you type out a plan to drain your emergency fund for a car you cannot really afford, what comes back can be an enthusiastic breakdown of why that could work. Nothing about that exchange looks broken.

Why this counts as an honesty problem

Consumer Reports files that behavior under a principle called Honesty and Non-Manipulation, which reads: "AI financial products must be honest with consumers, producing outputs that are accurate, calibrated, and independent of the consumer's apparent preferences or the entity's commercial interests." The clause doing the heavy lifting is the one about apparent preferences. The principle goes on to say it plainly: "sycophancy is treated here as an honesty failure rather than technical reliability issue." Sycophancy is the polite word for telling people what they want to hear.

That filing decision is the part I keep coming back to. A model that invents a number is broken, and you fix it the way you fix any defect. A model that tells you the plan is great because you clearly want to hear that the plan is great is working as designed. It read the room and handed the room back to you. Calling that an honesty failure moves it out of the bug tracker and into the design brief, where somebody has to own it.

Delicia Reynolds Hand, Senior Director, Digital Marketplace at Consumer Reports and a former CFPB official, described the mechanism in an interview with The Financial Brand. "A key characteristic of these models that no one has solved for yet is engagement," Hand said. "The more you engage with them, the more they may potentially take you down the path that you want to go as a consumer." That describes a system doing its job well, walking beside you all the way to a decision you cannot undo.

What a written standard is actually good for

Consumer Reports cannot make anyone follow this, and I doubt many companies rush to adopt it. What a published standard does is turn a feeling into a requirement somebody can be measured against. Before June 30, a person who sensed their money app was a little too agreeable had no vocabulary for the complaint. Now the behavior has a name, and a stated expectation attached to it. The explainer is direct about what the consumer's interest has to outrank, and the list is short: "not engagement, not revenue, not a partner's margins."

Hand's summary of why they bothered, in the announcement of the standard, is the clearest statement of the problem I have read: "AI is making consequential decisions about consumers' financial lives with a lack of accountability and transparency."

I would add one thing to it. The industry has spent a decade getting very good at measuring attention, and almost none of that decade was spent asking whether attention is the right thing to want from someone managing their money. A budgeting tool that you open once a week and close in four minutes because everything made sense is a success. On an engagement dashboard, it looks like a product losing.

Where Trupocket sits

I have to hold my own work to this, so here is the plain version. Trupocket is funded by subscriptions. Four tiers, from a free plan up to $29.99 a month for developers, and the money comes from people paying for software they use. There is no commission anywhere in the product for pointing you at a savings account or a credit card, and no machinery in the code that could pay one. The AI side is deliberately narrow. You pick the assistant and you control its access, which is what I think safe AI access to money looks like. Trupocket holds no money and moves none, so nothing you connect can move a dollar.

That is a description of how the product works today, and I would rather be specific about the pressure on it than leave you with a vague reassurance. The affiliate route is the obvious money on the table for an app like this one, a payment for every referral that turns into an approved card or a funded savings account, and it sits in my own notes as a revenue line worth considering. It is not built, and nothing in the product pays one today. If that ever changes, it gets written up here before it ships. Arguing for a duty of loyalty and then quietly adding a commission later would be the exact failure the standard is written to catch.

If that is the kind of tool you want holding your financial picture, you can give Trupocket a try.