In the dialogue between government representatives and the largest technology corporations, an agreement has been reached on launch platforms for implementing sovereign neural networks. Education and the public services sector have been recognized as priority areas. This information was confirmed by the Digital Economy ANO.

The concepts of sovereign and national models emerged thanks to a recently adopted law that regulates the development of artificial intelligence within the country. Now that the legal framework has been established, the stage of practical implementation has arrived.

Participants in the dialogue on the future of neural networks

Discussions between the industry regulator and leading technology giants took place at the junction of July and August. The artificial intelligence alliance joined the dialogue. The list of key participants in the negotiations includes the following companies:

  • Yandex
  • Sber
  • VK
  • Rostelecom
  • Rosatom

"[As a result] the authorities, in dialogue with business, have identified the first industries for priority implementation of sovereign and national AI models. These are education and the provision of public services," clarified the Digital Economy ANO.

The focus on these areas guarantees developers high demand from the state. Thus, creators of models with confirmed status will find applications for their digital products faster.

New rules for registration and evaluation of AI

A significant part of the discussion was devoted to mechanisms for assigning or revoking the special status of neural networks. According to representatives of the ANO, the bureaucratic procedure will become much more convenient for business. Programmers will no longer need to file a fresh application just to add missing information.

Also, the relevant departments decided to completely abandon excessive barriers. Key relaxations for neural network creators:

  • Cancellation of mandatory reports on training on state system databases.
  • Permission to rent computing power instead of mandatory ownership of infrastructure.
  • Exclusion of strict requirements for entering data centers into special registries.

In addition, officials are actively working on an algorithm for rapid updating of software products.

"Given the speed of their development, a mechanism is being discussed for linking status to a specific version with the possibility of simplified confirmation when releasing subsequent versions within the same product line," explained the organization.

The format for checking algorithms for compliance with laws and traditional values has also undergone changes. Soon, several accredited institutions will begin conducting such expert reviews. The auditing organizations themselves will regularly undergo strict audits by supervisory authorities. However, certain infrastructure restrictions will continue to apply.

My analysis: Choosing education and public services as pilot zones is a pragmatic step. Here, the state acts as both the customer and the guarantor of demand, which reduces risks for developers. However, success will depend on how quickly infrastructure restrictions can be overcome and real competition among accredited models can be ensured, rather than formal compliance with bureaucratic requirements.