AI models in Russia will be required to undergo a "spirituality" check for access to government procurement.
The Russian artificial intelligence market is on the brink of tectonic changes. This is not just about a technical regulation, but about introducing a fundamentally new filter: developers claiming the status of a "sovereign" or "national" model will have to prove that their algorithms comply not only with the letter of the law, but also with certain "traditional spiritual and moral values." Discussions of this scheme took place in late July and early August at the platform of the ANO "Digital Economy."
Formally, the framework law on AI came into force on July 26, but it only sets the contours. The key rules of the game will be defined by bylaws, and it is around them that the main struggle is currently unfolding. My analysis shows: we are witnessing an attempt by the state to create a precedent where compliance with ideological guidelines becomes a mandatory condition for access to significant budget funds and state support.
How the examination of algorithms will be structured
The verification mechanics look as follows. A developer submits an application to an accredited testing laboratory, providing a description of the architecture and information about request filtering mechanisms. The laboratory, in turn, conducts an examination: it runs the model through sets of tests (benchmarks) and checks its resistance to prompt injections — attempts to bypass internal restrictions through malicious instructions in the input data.
After that, the conclusion is sent to the government commission on digitalization. A positive verdict opens access to preferences in public procurement. Notably, they plan to check not all models, but only those claiming high status. The first areas of implementation will be education and public services.
Who will deliver the verdict
Here lies the main intrigue. Instead of a single appointed regulator, several laboratories that have passed verification by a government body will be able to conduct the examination. The scheme is two-stage: first, the developer tests the model itself using a general risk-oriented methodology, and then an accredited laboratory evaluates these findings and may selectively re-check the system's resilience. The final word will remain with the Ministry of Digital Development, while security issues for government systems will be controlled by the FSB and FSTEC.
The market participants I interviewed agree that the main risk is the "evaluative nature" of the criteria, which creates grounds for selective application. Alexey Borshchov, business architect of the Avandok ecosystem, rightly notes: categories such as civic consciousness or historical memory cannot be directly translated into a technical specification. The only possible common denominator is procedural: a unified report format, a set of provocative queries, and reproducibility of results.
The Institute for System Programming of the Russian Academy of Sciences is being considered as a potential operator of the checks. However, the package of bylaws is still undergoing interdepartmental coordination, and it is premature to talk about the final structure.
My expert assessment: The introduction of "spiritual censorship" for AI is a dual signal for the market. On the one hand, formalizing the procedure and accrediting laboratories is a step toward industry maturity. On the other hand, vague formulations create the risk of turning certification into a tool of administrative pressure. Investors and developers should closely monitor the final version of the regulation: the extent to which benchmarks and appeal mechanisms are transparent will determine whether this system becomes an incentive for development or a new barrier to innovation.