- Статьи
- Economy
- Frenzied algorithm: the Russian Academy of Sciences named the top vulnerabilities of the introduction of AI in the financial sector
Frenzied algorithm: the Russian Academy of Sciences named the top vulnerabilities of the introduction of AI in the financial sector
Market manipulation, an increase in cyber attacks, discrimination against borrowers and the devaluation of the workforce — the Russian Academy of Sciences warned of the main threats to the introduction of AI in the financial sector. At the same time, more than 70% of banks and insurance companies already use neural networks. They often use the same models and data, which is why algorithms make similar decisions and move markets in the same direction. Why credit institutions are still convinced that technology benefits more than harms, and how it will change the economy — in the material of Izvestia.
The main risks of using AI in the financial sector
Today, more than 70% of financial organizations use artificial intelligence. It is most often used in payments, credit analysis, insurance, and asset management. In all these areas, AI increases efficiency by reducing costs, ensuring regulatory compliance, detecting fraud, and improving customer service.
According to the McKinsey Global Institute, only in the global banking sector, the introduction of AI can annually bring 3-5% of the total revenue of the industry. Such data is provided by Dmitry Kochergin, Doctor of Economics at the Institute of Economics of the Russian Academy of Sciences, in the article "The main directions of using artificial intelligence in the financial sector."
However, the massive introduction of neural networks is accompanied by serious risks. The threat of cyber attacks is increasing. Generative models expand the capabilities of attackers to create phishing emails, malware, and hijack user devices. This can lead to data theft, extortion and fraud, warns the author of the study.
Another threat source is data poisoning attacks. Attackers can interfere with the arrays on which the language models of financial organizations are trained in order to change their behavior and disrupt the operation of systems. As more and more applications use data created by the models themselves, the operational risks for the financial sector are increasing.
A separate problem is the bias of algorithms. According to the author, models can reproduce and amplify distortions in data, which leads to discrimination in insurance and lending, and restrictions on access to financial services for vulnerable groups.
In addition, in conditions of strict data protection standards, the use of AI increases legal risks, especially due to the propensity for "hallucinations" of neural networks, Dmitry Kochergin's article notes.
Significant threats are also associated with the dependence of the market on a limited number of suppliers of language models. High development costs and data concentration lead to an oligopoly: several companies control the entire market, Dmitry Kochergin emphasized. Any failure or attack on them creates risks for all users. At the same time, even their own banking neural networks are usually based on the same technologies and behave in a similar way.
That is, different banks and investment companies use the same algorithms, which begin to make similar decisions automatically. As a result, players simultaneously buy or sell the same assets, move prices in the same direction, and actually act as a single coordinated market participant. There is no formal collusion, but the effect on the market is price distortion, increased volatility, and the risk of manipulation.
Additional risks are associated with the activities of AI agents capable of increasing imbalances and instability.
Leonid Delitsyn, an analyst at Finam Financial Group, also pointed out the risk of "poisoning" the training data when malicious code penetrates popular services and then spreads through AI recommendations.
Natalia Milchakova, a leading analyst at Freedom Finance Global, noted high infrastructure costs, the risks of wage inequality, rising cybersecurity costs and staff cuts, which could increase competition and drive small players out of the market.
In a positive scenario, with controlled implementation, AI is able to increase productivity, support economic growth, and slow down inflation. In the negative case, with spontaneous automation, there may be a devaluation of the labor force, an increase in defaults, an acceleration in price growth and a decrease in tax revenues. These factors can undermine financial stability, concludes the author of the study.
How financial organizations use AI
At the request of Izvestia, the Central Bank recommended reading the regulator's report on the use of AI in the financial market. According to the Central Bank, every second bank used neural networks in the second half of 2025. More than 80% of companies have implemented AI in insurance, and over 70% of professional stock market participants.
The editors asked the largest banks to tell them about their experience. VTB reported that they are actively using AI for credit scoring and underwriting, behavioral analysis, personalized offers, computer vision and automation of routine operations.
At Bank Sinara, the technology is used both in customer service and in the back office (internal departments of the company that do not interact directly with customers), in chatbots, voice assistants, anomaly analysis, data decryption, and even in design. At the same time, AI is still mainly involved in non-critical processes (which do not directly affect the client or the financial situation of the organization), said Alexey Averin, director of digital channel development at the company.
Novikom stated that they are carefully studying the market experience and consider concerns about risks to financial stability to be exaggerated. Dmitry Gritskevich, Head of Banking and Financial Market Analysis at PSB, noted that critical processes will continue to be duplicated by humans, and the scale of AI adoption is comparable to the spread of the Internet in the 1990s and 2000s.
VTB considers behavioral factors, rather than technical ones, to be the main risk — excessive trust in algorithms. Generative models can confidently formulate erroneous conclusions, and users tend to transfer responsibility for decisions to them, the bank stressed.
The All-Russian Union of Insurers reported that most insurance market participants test AI, but rarely trust it to make decisions. According to Gleb Yakovlev, Vice President of the VSS, technologies are effective in analyzing documents, but their large-scale implementation does not always reduce costs and improve the quality of service.
How AI helps banks
Financial market participants note that they still see neural networks as more useful than threats. VTB's key advantages include hyperpersonalization of services: AI analyzes hundreds of parameters and allows you to move from segments to individual offers. Another positive aspect is increased efficiency by automating routine processes and freeing up resources for more complex tasks. Technologies make it possible to create new services based on computer vision and robots, the bank added.
Novikom's advantages include reducing the number of errors, optimizing resources, multitasking, and working in 24/7 mode. The PSB noted that the models help recognize documents and voices, structure data packets, analyze transactions, and improve personalization accuracy. For developers, AI accelerates the release of new products through automatic code generation and error analysis.
Due to the widespread introduction of AI, the financial systems of many countries will change significantly, Dmitry Kochergin told Izvestia. The transformation will affect the architecture and infrastructure of the market, from the development of automated "smart" capital markets to the strengthening of the role of digital currencies of central banks and the decentralization of financial services. Business models will also change: some mass professions will disappear, autonomous AI agents and superapplications will appear, and product personalization and differentiation will increase.
According to the author of the study, supervision and regulation can also be reformed — with the transition to real-time control and application of monetary policy instruments and the adoption of global standards in the field of AI. The expert warned that such changes would increase system "fragility" and create a new generation of cyber threats.
"In general, financial systems will become more efficient and convenient, but probably less fair and transparent," said Dmitry Kochergin.
VTB expects that the massive adoption of AI will lead to two key shifts. The first is that an era will come when each customer will become a separate segment, and banks will be able to offer fully customized conditions in real time. The second is deep technological and cultural integration, including round-the-clock offices with robots and the development of both national AI tools and global models.
In the future, the financial system will become more automated and personalized: routine operations will switch to algorithms, and people will focus on strategic and ethical decisions, said Konstantin Kuchugurin, Technical Director of Compare. For customers, this means faster and more affordable services, but also higher requirements for digital literacy.
Переведено сервисом «Яндекс Переводчик»
