Russia wants to create its own "sovereign" artificial intelligence. It should work on Russian infrastructure and correspond to "traditional spiritual and moral values."
Russia wants to create its own "sovereign" artificial intelligence. It should work on Russian infrastructure and correspond to "traditional spiritual and moral values."
This was reported by the Center for Countering Disinformation.
Presumably, this is an attempt to create a large language model. The criteria are being developed by the Russian Ministry of Digital Affairs together with the FSB and the Federal Service for Technical and Export Control. AI will be tested for compliance with such values proclaimed in Russia as patriotism, "service to the Fatherland", a strong family, collectivism, the unity of the peoples of Russia, the priority of the spiritual over the material, and "historical memory".
Special benchmarks are also being prepared for AI — sets of questions and tasks that can be used to evaluate the model's behavior. In addition, during testing, the model will be deliberately tried to deviate from the "correct" course. The laboratories will use prompt injections — cleverly formulated queries that can force the AI to bypass its own limitations. This will check whether the model will start giving answers that the state considers undesirable as soon as the user finds the right way to ask it.
Previously, dev.ua wrote about a similar approach in China: AI models there must comply with "core socialist values." In particular, the updated DeepSeek R1 began to censor politically sensitive topics more often and reproduce the official position of the Chinese authorities.
