Financial News: Interview with Mikhail Mamuta for the journal “Banking Review”.

The financial market is moving away from crude impositions toward subtle algorithmic nudges. In an interview with B.O., Mikhail Mamuta, Deputy Chairman of the Bank of Russia, spoke about how to distinguish financial advisor advice and recommendations from manipulation, whether AI can be more ethical than humans, and where the advisor’s help ends and the investor’s own responsibility begins.

From rough imposition to native navigation

Pavel Samiev: Mikhail Valeryevich, let’s talk about the practice of client navigation in the financial market from the perspective of both financial organizations and advisors, intermediaries. Can we define criteria, highlight useful and harmful practices of native navigation, pushing the client towards decisions? If this is beneficial to financial organizations, can it also be beneficial to the client? Or is this a conflict of interest that needs to be addressed?

Mikhail Mamuta: The practice of client navigation is directly related to how the sales system developed in the financial market. Five or six years ago, financial organizations simply sold investment life insurance to clients under the guise of deposits and considered this normal. Then the regulator was granted the authority to suspend the sale of a product or service if it violated consumer rights, after which misselling practically disappeared. However, there was still some coercion: when you go for a loan, they sell you three insurance policies along with it, only one of which is required to obtain the loan. Thanks to deputies and senators, we managed to introduce legislative changes that helped practically eradicate this practice.

But financial products need to be sold somehow, and a shift came from crude deception or coercion to more subtle mechanics of gentle nudging. The person thinks they are making the decision independently, but in fact, they are skillfully guided toward it. It’s like with supermarkets: many people don’t fully understand why they came for bread but ended up buying a full basket of products they basically don’t need. Because the goods were arranged on the shelves and accompanied by such discounts that you thought, “How could I not take this when it’s cheap.” Although maybe it isn’t any cheaper than in the neighboring supermarket without discounts. The seller’s desire to secure market leadership will never disappear.

How do we feel about this?

We are not at all opposed to the very idea of navigation, when a person is helped along a certain path to make the best choice. But we are against highlighting only the positive qualities of the product on this path while hiding the negatives in the shadows.

The price of “stars”: how to restore honesty in deposits

Pavel Samiev: So, the client only sees the tip of the iceberg?

Mikhail Mamuta: Even in the past year, 2025, we received quite a few complaints regarding deposits, when people said: “I was promised a deposit with an elevated interest rate, but it turned out that to receive it, you need to have a minimum required balance on the account, or open a brokerage account, or make card purchases for a certain amount – overall, fulfill additional conditions.” As a result, the consumer is disappointed because they received less than expected. And this is a serious problem.

We had long internal discussions within the Bank of Russia about what to do with this. Initially, a strict model was proposed – to standardize the main financial products. But even we didn’t like this option – it’s not quite a market-based approach. In the end, the idea was born to create an AI assistant that would use the capabilities of LLM to highlight the key features of the product and the terms of the contract. Thus, the consumer will have all the necessary inputs for making a decision.

We tested this model and see that it handles the task assigned to it quite well, it was approved by people and, strangely enough, did not cause rejection among banks. Banks understood: they can have not a negative conversion, but a positive one, because people better understand the essence of the contract. And here is what is important: partners who really want the client to stay with them for a long time and not be disappointed joined this pilot project. For them, this model is very profitable. Because you can deceive someone once or twice, but the third time the person will turn around and leave. And if you want them to be with you throughout their 30 years of active life and remain satisfied, you cannot deceive their expectations.

Standard for a neutral advisor

Pavel Samiev: Could this tool become a common standard for the market? Might it happen that unscrupulous players gain an advantage by abandoning fair navigation?

Mikhail Mamuta: In my opinion, a scenario in which the use of such an assistant could be established by a certain standard is quite likely. This would help reduce the risks of unfair competition, and we are always against it. Under such conditions, for some time, the advantage remains with those who act not in the interests of customers, but only pursue their own benefit. In the end, consumers will sort it out, but it may take more than one year and lead to disappointed expectations. It is better to have clear rules in the market.

We must be confident that financial organizations will use the AI assistant in good faith, rather than turning it into a biased sales service.

Neutrality, ethics – that’s what interests us. These principles can also be enshrined in a standard.

Standards are developed together with banks, which play a fundamentally important role in shaping new requirements.

At this stage, this path seems optimal to me.

Machine empathy: why AI begins to “flatter”

Pavel Samiev: Do you observe that clients tend to trust artificial intelligence in various forms more than a human—manager, advisor—when navigating financial instruments? Is there a problem with this? Will it affect the quality of sales and navigation? How can the operation of artificial intelligence be regulated?

Mikhail Mamuta: This is a very good question. It concerns one of the most important values not only in the financial market but in life in general – trust. Trust may seem like an ephemeral concept, but in practice, it is always based on a certain amount of measurable indicators.

Three years ago, many had the impression that the hype around AI would pass. Today, my feeling is that this is a new reality, related to the possibility of creating systems that are gradually approaching how the human brain works.

According to several international scientific studies, the latest AI models formally already demonstrate signs of what is called AGI – Artificial General Intelligence (strong artificial intelligence). The problem is that it will be psychologically difficult for us to admit this – our anthropocentrism hinders us from agreeing that something resembling human intelligence has appeared. But where is all this leading? Trust in language models grows as their ability to provide relevant advice increases. This is highlighted by recent research. The invasion of language models into everyday life is already happening, and it will also occur in the financial market.

A neutral, impartial AI assistant may inspire more trust among consumers than a salesperson at a financial organization, who can sometimes be caught being biased.

But it is important to remember that the decision always remains with the person themselves; no one has absolved them of responsibility for their actions.

As for the prospects of AI regulation, we are currently proceeding with methodological recommendations, providing guidelines on how to build safe client-oriented services.

Pavel Samiev: I’ve come across in some survey that people increasingly begin to trust artificial intelligence because they believe that, unlike a human, it has no conflicts of interest when making recommendations, that it won’t deliberately distort information or impose anything.

Mikhail Mamuta: People trust those they like more. If a person communicates respectfully and evokes sympathy, we unconsciously trust them more than someone who might say fairer things but does not show empathy towards us. The latter will cause unconscious rejection; these are characteristics of human nature.

The problem is that the more advanced LLM models are increasingly adapting to human expectations. We see that they start to “flatter.” They don’t just answer the question; they say things like, “Wow, you’re so smart, few people ask such well-formulated questions,” “You expressed yourself really well,” “I was just thinking about the same thing,” “You’re doing great.” A person involuntarily starts to be drawn to such an advisor: it feels good to be praised, to be valued so highly. And this is where the main risk lies: up to what point does a person realize they are being played along with, and from what moment do they lose control in their communication with their “interlocutor”.

The second important aspect is that a model fine-tuned to the needs or interests of a specific organization is, unfortunately, no longer science fiction. Such models can conscientiously compare products from different companies, but for some reason tend to favor the one that trained them.

The boundaries of permissibility for the digital mind

Pavel Samiev: So, does this require standardization and special stipulations?

Mikhail Mamuta: The state must define the principles of ethics and neutrality that AI must not cross. It should be established what disclaimers, disclosures, and explanations are provided to a person when navigating using artificial intelligence.

Otherwise, if language models gain superintelligence, there will be a risk that we will stop understanding whether what is happening is under our control or even be aware of what is happening. But until then, I think we are still quite far away.

Pavel Samiev, “Banking Review”