The expert spoke about the concerns of businesses when using AI

The volume of confidential information that employees of Russian companies transmit to public neural networks has increased by more than 30 times in 2025. This was stated to Izvestia by Sergey Golitsyn, head of T1 AI. Requests for language models include commercial offers, contracts, financial statements, personal data of clients and business correspondence, he added.

By action code: scammers use the ChatGPT viral trend to collect data from Russians

We are talking about a flash mob with the disclosure of your personal information — how not to become a victim

"The main problem is the lack of control over data transmission to AI systems. At the same time, 60% of organizations do not have a clear policy on the use of AI tools and large language models (LLM). Scenarios where developers use public LLM interfaces to work with code are particularly dangerous: fragments of internal logic and sensitive business information may leak into external services," the expert explained.

According to Sergey Golitsyn, software created exclusively with the help of AI contains 15 times more vulnerabilities than human-written software. Among the most frequent are the lack of input filtering and incorrect authorization. Another threat is related to agent systems: about 62% of companies are testing them today. If compromised, such systems give attackers access to the entire IT circuit of the organization.

To solve the problem, the expert recommends deploying AI tools inside your own IT perimeter and using proven data libraries. For critical industries (financial sector, government agencies), a hybrid model can be the best solution: a local hardware and software complex for working with sensitive data and a trusted cloud for scalable computing.

Read more in the Izvestia article:

People and leaks: why domestic business refuses to implement AI