Rostelecom plans to create a corporate firewall for language models for its own needs, informs the Telegram channel "Telekommunalka", registered with Roskomnadzor, citing industry sources. The gateway will check employee requests and neural network responses, and applications from developers are accepted until October 19.
At least 1,500 generative artificial intelligence (genAI) solutions are already operating within Rostelecom's perimeter. Targeted checks are insufficient for such a volume, so a unified control layer is needed. According to the technical specifications, the system will be tested on Rostelecom's own infrastructure at the corporate center in Noginsk. The gateway must recognize attempts to bypass model restrictions, inject malicious instructions, extract system settings, or covertly upload confidential data through leading questions.
The firewall will be integrated with the company's internal services: Active Directory for authorization, SIEM/SOC for transmitting security events, and DLP for leakage control. The software code must be included in the Russian Ministry of Digital Development's register of Russian software. This confirms that the customer is the operator itself, protecting its corporate secrets, and not a third-party client. This is an internal project, not the purchase of a ready-made product for the market.
For now, this is a request for information, not a tender. If the pilot reaches implementation, it will become one of the largest examples in Russia of protecting generative AI within a single company. Rostelecom is effectively purchasing development "for itself" to prevent data leaks through its 1,500 AI services. Demand for such systems will grow along with the number of LLM applications and new neural network threats.