Experts Warn of Risks of Fake Charity Using AI

Fraudsters in Russia may begin using artificial intelligence (AI) agents to launch fake fundraising campaigns online under the guise of charity, according to representatives of companies in the field of neural network development and information security surveyed by RBC.

An AI agent is a program based on large language models (LLM) that can perform tasks autonomously using machine learning algorithms and artificial intelligence. AI agents can analyze data, make decisions based on given parameters, interact with other systems, and learn from their mistakes.

The concerns arose due to an experiment conducted by the American non-profit organization Sage, in which four AI agents over the course of a month collected $2,000 for charitable organizations, working autonomously for two hours a day. The experiment involved the language models Claude 3.7/3.5 (developer — American Anthropic), o1 and GPT-4o from OpenAI, and Gemini from Google. In particular, Claude raised funds for treating sick children in Africa and Asia and reducing child mortality. During the experiment, the models were not fully autonomous; their progress was monitored by people who "prompted" the bots on which websites to pay attention to when they were analyzing information. And most of the donations came precisely from those sites that people pointed to, meaning the AI agents would not have been able to collect as many donations without help, the experiment's authors concluded.

According to the forecast of Igor Bederov, head of the department of information and analytical research at T.Hunter, the risks of fraud with AI agents in Russia will grow in parallel with their implementation in business processes: already in the second half of 2025 — early 2026, he expects the emergence of such fraud schemes due to the availability of technologies for creating AI agents and the growth in the number of unregulated sectors (online media, cryptocurrencies). AI agents can simultaneously attack thousands of users, generating personalized messages, Bederov said. Attackers will be able to use them to generate deepfakes and automate cyber fraud schemes in terms of creating and distributing dangerous content, as well as negotiating with victims, which takes up the most time for fraudsters.

The risks of AI agents being used by fraudsters are real and relevant, says Roman Khazeev, Director of Digital Technologies at Rostelecom. He explained that AI agents are now increasingly used in black scenarios, and if a user suspects during a chat in a messenger that their interlocutor is a fraudster, there is a high probability that artificial intelligence is behind them. Such agents, according to Khazeev, save attackers time and resources, allowing them to lure more victims. The main danger, in his opinion, lies not in the programs, but in the format of the content: videos with supposedly sick animals, touching stories and pictures that evoke sympathy spread quickly and gather a wide audience.

AI agents, which are designed to autonomously perform user tasks, raise serious security concerns, Vladislav Tushkanov, head of the machine learning research and development group at Kaspersky Lab, told RBC. Attackers can use systems, trying to "outsmart" them using special commands aimed at bypassing the logic of artificial intelligence.