Russian neural networks may be allowed to train on state data

Rules for artificial intelligence in Russia may change. A new version of the AI regulation bill proposed allowing domestic neural networks to train on state data. The publication "Kommersant" reviewed the document.

According to the new version of the bill, national and sovereign AI models will be able to access state data for training after approval from the FSB of Russia and the FSTEC of Russia. The document also introduces responsibility for AI owners and operators related to ensuring the information security of neural networks.

Another important change concerns the status of trusted AI. A model can only be considered trusted after inclusion in a special register, the procedure for which will be determined by the Russian government.

In addition, the government will be able to establish cases of mandatory use of national and sovereign AI models in the banking sector – in agreement with the Central Bank.

Previously, inclusion in the register was only envisioned for models used in critical information infrastructure facilities and state information systems.

The head of the Russian Ministry of Digital Development, Maksut Shadaev, emphasized that Russia gives maximum priority to the development of neural networks and AI technologies. The ministry reported that the bill is still being finalized with the industry and interested agencies. They clarified that the criteria for including models in the trusted AI register will be determined by separate regulatory acts.

Market participants are already discussing possible innovations. T-Tekhnologii noted that the document uses the concepts of "sovereign" and "national" models, but their exact meaning has not yet been disclosed. The Big Data Association emphasized that regulation should not create additional barriers for developers.