K2 Cyber Security and Kaspersky published a study on AI security in Russia. According to the data, 38% of Russian corporations develop AI solutions for internal processes. However, only 14% of companies are familiar with the MLSecOps methodology, while 46% are unaware of it and 40% have only superficial knowledge.
The study highlights gaps in AI infrastructure management. Only 32% of organizations maintain a centralized model registry, and 43% have a centralized dataset catalog. While 61% of companies implement model version control, the lack of unified operational contours indicates fragmented responsibilities between development, data management, and security teams.
Basic security practices are partially adopted, with 59% of companies implementing access and delivery controls and 46% securing their datasets. However, advanced monitoring remains rare. Only 15% of organizations have implemented drift monitoring, with another 15% currently piloting the technology, leaving 70% without any such oversight.
Experts note that the AI security market is still emerging. Risks are currently distributed across various functions, including development security and infrastructure monitoring. As threats evolve from simple data leaks to malicious code in AI agents, vendors are focusing on new tools like AI Firewalls and employee training programs.