Spiritual screening for AI: Russia introduces mandatory testing of models for traditional values
Russia's artificial intelligence market is on the verge of tectonic shifts. Developers claiming the status of a "sovereign" or "national" model will be required to confirm that their products comply not only with legislation but also with certain "traditional spiritual and moral values." This procedure, actively discussed in late July and early August on the platform of the ANO "Digital Economy," radically changes the rules of the game for all major industry players.
Formally, the AI law came into force on July 26, but it is framework in nature. Key mechanisms, including the notorious "spirituality" check, will be spelled out in by-laws that are currently in the final stage of approval.
How will the new system work?
My analysis shows that the regulator is moving toward creating a two-tier certification system. The law introduces two special statuses for large foundational models — "sovereign" and "national." To obtain one of them, a developer will have to prove that their AI does not contradict traditional values.
Notably, there are no plans to draft a separate act listing these values — the list is already enshrined in Presidential Decree No. 809. Instead, special benchmarks — sets of tests for evaluating models — will be developed jointly with businesses.
The procedure will look as follows: a developer submits an application, an architecture description, and data on request-filtering mechanisms to an accredited laboratory. Experts then check the model's responses on benchmarks and test its resilience to prompt injections — the so-called "AI hacks" where a malicious instruction is disguised within input data.
A positive expert opinion will serve as a pass to state preferences, including priority access to government procurement. However, not all models on the market will be checked, only those vying for high status. According to my data, the first implementation areas will be education and government services.
Who will deliver the verdict?
The key question is who will be granted the right to conduct the review. IT companies have agreed that it will not be a single appointed organization, but rather several laboratories that have passed verification by a state body. The scheme is being built in two steps: first, the developer independently tests the model using a common risk-based methodology, and then an accredited laboratory evaluates these findings and may selectively verify the model's resilience.
The final decision on status will remain with the Ministry of Digital Development. Security requirements for state systems will be checked by the FSB and FSTEC. The Institute for System Programming of the Russian Academy of Sciences may also claim the role of reviewer.
However, this initiative has serious pitfalls. Alexey Borshchov, business architect of AI projects in the Avandok ecosystem, rightly notes that abstract categories such as civic consciousness, justice, and 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.
Dmitry Galantsev, managing partner of the law firm "Propositum," warns of the main risk — the "evaluative nature" of the criteria, which opens the door to selective application. He proposes a more robust scheme: an accrediting body, several competing laboratories, and an appeals board outside the industry, plus a public registry of conclusions and certificate validity periods.
My verdict: this initiative is a vivid example of how regulation tries to ride the technology wave. On the one hand, creating a transparent AI certification system is a step in the right direction for security. On the other, attempting to measure the "spirituality" of an algorithm through benchmarks looks like a bureaucratic oxymoron that could lead either to a formal approach or to the creation of a new tool for pressuring the market. Investors and developers should closely monitor the final version of the by-laws — that is where it will be decided whether this system becomes a working mechanism or yet another decoration.