AI models will undergo a spirituality check: access to government procurement will become a privilege
Russian artificial intelligence developers will face a new barrier: their models will have to confirm compliance not only with legislation, but also with traditional spiritual and moral values. The procedure is expected to be carried out in independent testing laboratories. This initiative was actively discussed in late July and early August at the platform of the ANO "Digital Economy".
How the verification will be structured
The federal law on AI, which came into force on July 26, is framework in nature. Key rules, as I expected, will be established by subordinate legislation. The law introduces two special statuses for large fundamental models — "sovereign" and "national". To obtain one of them, a developer must prove that their product complies with legal norms and traditional values.
No separate document with a list of values is planned yet — the list is already fixed in Presidential Decree No. 809. Instead, businesses, together with regulators, will develop special benchmarks — sets of tests for evaluating models. The procedure will look like this: a developer submits an application to a laboratory, attaching a description of the architecture and data on request filtering mechanisms. Experts then check the model's responses against the benchmarks and test its resistance to prompt injections — the so-called AI hacking, when a malicious instruction is disguised in input data.
A compliance conclusion is sent to the government commission on digitalization. A positive verdict opens access to state support, including preferences in public procurement.
It is important to emphasize: no one intends to check all models on the market. This concerns only developments seeking the status of sovereign or national. The first areas of implementation will be education and public services.
Who will issue the verdict
The scheme involves a two-level system. First, the developer independently tests the model using a general risk-based methodology. Then an accredited laboratory evaluates their findings and may selectively verify the model's resilience. The final decision on status remains with the Ministry of Digital Development. Security requirements for state information systems will be controlled by the FSB and FSTEC. The largest developers already have tools for blocking prohibited requests, so the ministry expects to rely on market experience.
Industry experts agree that a substantive interpretation of traditional values is a non-trivial task. Civic consciousness, justice, and historical memory cannot be directly translated into a technical specification. Only a procedural approach is realistic: a unified report format, a set of provocative requests, and reproducibility of results. The key risk is the "evaluative nature" of the criteria, which could lead to selective application. A scheme with an accrediting body, several competing laboratories, and an appeals board outside the industry seems workable. It is also proposed to introduce certificate validity periods and a public registry of conclusions.
My analysis: This is a landmark step that transforms AI regulation from formal to substantive. However, without clear, measurable metrics for "spirituality," the system risks becoming a tool of bureaucratic bargaining. The market needs transparent benchmarks and independent laboratories, otherwise innovation will go underground, and only loyal players will receive statuses. Time will tell whether this initiative can become a model for other jurisdictions or remain a unique experiment.