Solar: Nearly 40% of Employee AI Interactions Contain Confidential Data

Solar Group analyzed the use of public AI services in large Russian companies, covering 12,000 employee interactions recorded during Solar Dozor DLP system pilots in the first half of 2026. In 38% of cases, these interactions contained confidential information: 41% involved source code and configurations, 30% personal, financial and other sensitive data, and 18% intellectual property. Development teams accounted for 43% of such events, and commercial departments for 26%. The analysis included 150 companies from the financial sector, industry, retail, telecom, IT, and government. Solar Dozor's product head Ali Gadzhiev noted that public AI is now a daily tool, but creates management risks, so companies need to regulate and control AI use. The data transmitted included code, documents, tables, contracts, customer information, and internal reports. Passwords, tokens, and API keys made up 11% of confidential interactions. A joint study by UCSS and Solar among 102 companies showed that 82.8% allow generative AI for technical staff, but 42.4% suspected AI-related leaks, and 8.1% confirmed incidents. The company recommends a two-level management approach: controlling access to AI services and monitoring the content of transmitted data, along with security awareness training for employees.