The Institute

Legal details

Autonomous Non-Profit Organization (ANO) “Scientific Research Institute (NII) of System Synthesis” (Russian short name: ANO “NII Sistemnogo Sinteza”), developing third-generation Information Security (IS, IS 3.0) for artificial intelligence (AI) systems.

Subject of work

The institute researches the engineering method of a managed transition: how autonomous AI systems can work under the oversight of organizations and the state, keeping their usefulness without breaking the order they are built into.

The field crosses three classical areas: formal verification of computing systems, the design of corporate and government compliance programs, and the methodology of governing autonomous multi-agent systems. The institute works where these areas meet and builds a unified discipline for them.

The institute’s name names its method: requirements that exclude each other are separated along different coordinates of the system rather than averaged. The procedure, the required output of each step, the conditions for refusal and the limits of the method are set out separately.

Three lines of work

The institute was registered on 12 March 2026 (the decision to found it dates from 9 December 2025) for the purposes of scientific research and the commercialization of its results. Its activity is built along three lines of work that run as a single research program. Its registered activity codes (OKVED) are 72.19, scientific research and development (main), and 62.01 and 62.09, software development and information-technology services.

Research

The research program studies the formal structure of the AI-led organization, deterministic autonomous systems and compliance by design. Its scientific novelty lies at the intersection of formal verification, organizational rules and governed autonomy.

Technology

Development of software components of the stack: orchestration of an autonomous corps, integration of Satisfiability Modulo Theories (SMT) verification, all components deployed inside the client’s perimeter.

Knowledge transfer

A book, analyses and talks for corporate clients and government agencies: the structure of an organization in which AI agents act, the requirements of Federal Law 187-FZ and Federal Service for Technical and Export Control (FSTEC) Order 117, formal methods in governing autonomous systems.

How the institute is organized

The institute is organized as a control system with feedback, following Stafford Beer’s Viable System Model. Its operational work was handed to computing machines from the start, and the scientific director keeps only what cannot be delegated: goals, values and legal responsibility.

The five functions of a viable organization, namely operations, coordination, control, intelligence and policy, are held by four autonomous working units and one person at the policy level. In every unit an executor does the work and an independent observer only reads and assesses it; the observer runs on a model from another developer, so their mistakes do not coincide.

Knowledge lives in a single body of rules and in logs rather than in employees’ heads, so the method does not depend on who sits at the keyboard. A program’s report of success is not accepted as done work: acceptance checks the live source on the server, the record in the database and the checksum of the delivered file. Operational execution runs as the VSM-MITR twelve-role electronic corporation across three hardware station pairs (GAMMA, BETTA, EPSILON) protected as a trade secret under Articles 1465 and 1262 of the Civil Code of the Russian Federation, with ten-minute model-agnostic provider switching and a four-hour technical escrow recovery contour. The full description with diagrams is on the page “How the institute is organized” (in Russian), and the commercial product suite is on the page “Products”.

Four autonomous units carry the work and one person holds the policy; in every unit an independent observer on another developer’s model assesses what the executor did, and acceptance checks the live source, the record and the checksum rather than a report.01 · FIVE FUNCTIONSFour units and one personpolicy · one personintelligencecontrolcoordinationresearchproductscontrolstandardsOperations, coordination, control, intelligenceand policy; goals and liability stay with a person.02 · WHO CHECKSAn executor and an observerexecutorthe workobserveranother developer’s modelThe observer runs on another developer’s model,so their mistakes do not coincide.03 · WHAT COUNTS AS DONESource, record, checksuma program’s reportlive sourcerecord in the databasechecksum of the fileA program’s report of success is not accepted;acceptance checks the server, database and file.Four autonomous units carry the work and one person holds the policy; in every unit an independent observer on another developer’s model assesses what the executor did, and acceptance checks the live source, the record and the checksum rather than a report.01 · FIVE FUNCTIONSFour units and one personpolicy · one personintelligencecontrolcoordinationresearchproductscontrolstandardsOperations, coordination, control, intelligenceand policy; goals and liability stay with a person.02 · WHO CHECKSAn executor and an observerexecutorthe workobserveranother developer’s modelThe observer runs on another developer’s model,so their mistakes do not coincide.03 · WHAT COUNTS AS DONESource, record, checksuma program’s reportlive sourcerecord in the databasechecksum of the fileA program’s report of success is not accepted;acceptance checks the server, database and file.
Four autonomous units carry the work and one person holds the policy; in every unit an independent observer on another developer’s model assesses what the executor did, and acceptance checks the live source, the record and the checksum rather than a report.

Related resources

The institute’s applied services, on whose corpora the method is tested. Commercial licensing and enterprise deployment operate through the dedicated operating company MITR Technology LLC while core intellectual property remains with the institute; the status of each service is given as of the date checked.

ServiceWhat it doesStatus as of 15 November 2026
SUDFACT (in Russian)court-practice analysis: 14,899 court acts with statistics down to the case cardrunning; corpus measured on 25 October 2026
State-duty calculator (in Russian)computes the state duty under the Tax Code of the Russian Federationrunning
FEDRESURS (in Russian)publication of mandatory notices in the Federal Registerrunning
“Juridika”corporate legal suite with local models inside the perimeter and Juridika Shield three-valued citation checkcorporate contour; pilot deployment

Open to cooperation

The institute works with three groups of partners: owners of critical infrastructure and regulated processes; investors who see autonomous systems as an asset class; academic and industry research organizations.

The institute is open to scientific cooperation with universities, industry academies and the research institutes of the Russian Academy of Sciences. Formats: joint research programs, expert councils, participation in dissertation work in related fields.

Direct correspondence is accepted at af@isslab.ai, while the contact form at isslab.ai/contact/ operates in Russian. Ongoing research notes are published in the institute’s Telegram channel “IS 3.0: AI agent security” at t.me/agiandhuman.

What has been done

Since 24 April 2026 MITR has been running: it checks the actions of AI agents before they run, and since 24 October all 20 of its contracts have been in enforce mode. Open packages for developers have been released, the INDRA (Invariant, Nondestructive, Deterministic Resilience Audit) tester has passed acceptance on an internal reference bench prior to external laboratory accreditation (page in Russian), the scientific director’s book Proven Action has been published, and two patent applications filed on 1 June and 22 September 2026 are under examination. Figures with dates of measurement and ways to check them are on the page Investors and partners.

Registration details and activity codes were verified on 15 November 2026 against the Unified State Register of Legal Entities. All legal acts and standards named on this page were checked against primary legislative registries.