Spiritual control for AI: Russia introduces mandatory checks of models for "traditional values"
The Russian artificial intelligence market is on the brink of tectonic shifts. Developers claiming the status of a "sovereign" or "national" model will soon be required to confirm not only technical compliance with legislation, but also adherence to traditional spiritual and moral values. This procedure will take place in accredited testing laboratories, and discussions about its parameters were held on the platform of the ANO "Digital Economy" in late July — early August.
How the new verification system works
On July 26, the framework law on AI came into force, but the key rules of the game still need to be spelled out in by-laws. Currently, the introduction of two official statuses for large fundamental models is being discussed — sovereign and national. To obtain one of them, the developer must undergo an examination for compliance with legislation and value orientations.
Notably, a separate act with a clear list of values is not yet planned — the list is already enshrined in Presidential Decree No. 809. Instead, special benchmarks will be developed jointly with businesses: sets of tests for checking models. The procedure will look like this: the developer submits an application with a description of the architecture and request filtering mechanisms, after which the laboratory checks the model's responses against the benchmarks and tests it for resistance to prompt injections — the so-called "AI hacking," when malicious instructions are hidden in input data.
A positive conclusion is sent to the digitalization commission, and it is this conclusion that opens access to state support and preferences in public procurement. At the same time, no one intends to check all models on the market — only those claiming a high status are in question. The first areas of implementation will be education and government services.
Who will deliver the verdict
Instead of a single authorized organization, it was decided to create a network of several accredited laboratories. The scheme is being built in two stages: first, the developer independently tests the model using a common risk-based methodology, and then an accredited laboratory evaluates the findings and may selectively verify the system's resilience. The final word will remain with the Ministry of Digital Development, and security requirements for government systems will be coordinated with the FSB and FSTEC.
However, experts are already pointing to serious risks. Alexey Borshchov, business architect of the AI ecosystem "Avando," rightly notes that categories such as civic consciousness, justice, and historical memory cannot be directly translated into a technical specification. The only workable option is a procedural approach: a unified reporting format, a set of provocative queries, and reproducibility of results. Dmitry Galantsev, managing partner of the bureau "Propositum," warns about the "evaluative nature" of the criteria, which could lead to selective application, and proposes introducing certificate validity periods and a public registry of conclusions.
My view: this is a landmark precedent for the entire industry. Formalizing "spirituality" in the form of benchmarks is an extremely complex task, and the market awaits not so much the law itself as its enforcement practice. If the criteria remain vague, we risk getting a tool not so much of regulation as of administrative pressure. Transparency of the procedure and a clear methodology are the only way to avoid chaos in this new reality.