Scientific Director

Aleksandr Fishchuk, scientific director of the Research Institute of System Synthesis, systems engineer

Aleksandr Fishchuk (ORCID 0009-0000-6549-6113)
Founder and scientific director of the Autonomous Non-Profit Organization (ANO) “Scientific Research Institute of System Synthesis” in Moscow, systems engineer, author of the book Proven Action, and architect of the MITR (Model-Independent Task Regulator) proven-action gateway (running since 24 April 2026) and INDRA (Invariant, Nondestructive, Deterministic Resilience Audit) bench. The professional path spans more than twenty years in regulated environments. Early work included internal control at a subsidiary of an oil company (2005–2006). Financial management at an international grain trader followed (2006–2008), and later practice since 2008 focused on crisis management, insolvency, and corporate-risk governance. At the institute the scientific director sets the goals of research on information security of the third generation (IS 3.0) for artificial intelligence (AI) systems, signs the rules of MITR and answers for them.

Education

  • 1997 to 2002 — Volgograd State Technical University: engineer-economist.
  • 2003 to 2006 — Institute of Professional Accountants of Russia: professional accountant (IFRS).
  • 2012 to 2013 — North-West Institute of Management, Russian Presidential Academy of National Economy and Public Administration (RANEPA): crisis management; insolvency practitioner (arbitration manager).
  • 2012 to 2015 — Financial University under the Government of the Russian Federation: law.

Professional career

  • 2026 to present — Founder and scientific director of the ANO Scientific Research Institute of System Synthesis.
  • 2016 to present — Legal practice in capital protection, director-liability defense, insolvency proceedings, and corporate tax security.
  • 2008 to 2016 — Head of consulting firms: business support, commercial litigation, support during tax audits, reorganization and liquidation of legal entities, insolvency proceedings.
  • 2006 to 2008 — international grain trading company W. J. Grain (WJ Grain), serving as chief accountant and chief financial officer.
  • 2005 to 2006 — oil production enterprise Lukoil-Nizhnevolzhskneft, serving as an engineer in the internal control department.
  • 2003 to 2005 — FiKON: external auditor.
  • 2002 to 2003 — Rosmashkom: engineer-economist.

Methodological position

Formally verifiable solutions take priority over behavioral ones, local deployment over the cloud, and the discipline of provenance over the optimization of metrics.

The methodological task is to separate the domains of applicability of different tools: the load-bearing loops of high-consequence systems require guarantees derived from formal axiomatics rather than empirical distributions.

In this logic the institute’s program complements the dominant artificial intelligence approach across classes of problems where the cost of an operational error exceeds the cost of detecting it before execution.

The thread of a professional career

More than twenty years of work in regulated environments with formally defined normative frameworks: internal control engineer at a subsidiary of an oil company, chief financial officer of an international grain trader, crisis management, tax practice, insolvency support, and corporate capital protection. What runs through this path is work with systems in which a decision is judged correct not by its result but by its compliance with a standard of proof, and in which an error costs far more than the gain from making it.

The Research Institute of System Synthesis was founded as a natural continuation of this line in the field of autonomous computing systems. The architectural principle of Compliance by Design, building conformity with institutional requirements into the axiomatics of a system before it executes, coincides in method with the principle of audit and legal work: the norm is checked when an operation is designed, not when its consequences arrive.

This coincidence is not an analogy. It is the basis for carrying the discipline of formal checking over from legal and financial practice into the design of autonomous agents.

The program’s working implementation, MITR, has been running since 24 April 2026 and checks the actions of the software agents that the program’s author uses in professional practice. The program’s architectural claims are therefore tested by daily operation rather than by a demonstration.

How professional experience relates to the subject of the program

The institute builds an operating environment in which an autonomous executor works under conditions similar to those of a regulated profession: a fixed normative environment, mandatory logging of actions, separation of functions by the principle of irreversibility, and keys to the target systems held by MITR rather than by the executor.

In substance this task coincides with that of internal audit, tax support and legal defense in regulated industries. The difference is that the classical disciplines apply formal checking to human decisions after the fact, while the institute’s program moves it into the execution path of an autonomous system and applies it before execution.

Most of the cases that make up legal and audit practice arise not from an actor’s bad intent but from a mismatch between the action taken and the normative environment the actor belongs to. If every professional decision were checked against the applicable legal rules before it was made, a considerable share of the later defensive, restorative and rescue procedures (tax disputes, insolvency proceedings, corporate conflicts) would have no grounds to arise.

The fundamental obstacle to such a regime for a human executor is the cognitive cost of constant normative checking: for most professional decisions the volume of applicable norms exceeds the practical limit of attention and working memory.

This limitation gives rise to the class of problems served by the classical retrospective disciplines. An autonomous computing system has no such limitation: checking an action against the axiomatics takes a few milliseconds and can run continuously at no cognitive cost.

In this logic an autonomous agent with formal verification of its decisions does not replace the classical retrospective disciplines. It supplies the part of their function that a human executor could never perform continuously.

A further mechanism seen in this practice is an executor repeating successful decisions from past experience after the normative environment has already changed.

By its economic nature professional thinking works through pattern matching: accumulated cases become templates applied to new situations. This mode is effective in a stable environment and fails systematically under accelerated regulatory drift: the actor builds a decision on a picture of a normative field that no longer exists.

Structurally this failure coincides with the hallucination of a large language model: a conclusion plausible in form that diverges from current reality, because generation relies on a distribution detached from the environment where it is applied. The only difference is the source of the distribution: professional experience for a person, the training corpus for a language model. The architectural answer is the same in both cases: an external check of the action against the current normative axiomatics before it is executed.

Experience with IFRS financial reporting, internal control at a subsidiary of an oil company, crisis management and legal practice provides the grounds for two research positions of the program.

The first is that formal verification of an autonomous agent’s actions is not scientific exotica but an infrastructural norm, missing from current practice for reasons of design rather than technology.

The second is that the regulatory framework, including Federal Law 187-FZ, Federal Service for Technical and Export Control (FSTEC) Order No. 117 (in force since 1 March 2026), and Russian national standards (GOST, GOST R 57580), sets the exact engineering specification by which autonomous systems should be built from the start. Under Article 53.1 of the Civil Code of the Russian Federation and Article 61.11 of Federal Law No. 127-FZ on Insolvency, personal liability of chief executives turns on documented proof of reasonable and good-faith oversight: an admission gate verified by 6,414 automated tests and a cryptographic receipt chain (5,094 records in the 25 October 2026 book baseline; 5,373 records across 55 completed epochs as of 15 November 2026) converts an uninsurable operational hazard into verifiable corporate evidence.

Publications

Cover of the book Proven Action (Russian edition)
Book · 2026 · 856 pages · in Russian
Proven Action. Power and Control over AI Agents: Strategy 2027–2035

Author: scientific director of the institute

For owners and chief executives: who actually controls an AI agent and who answers when it makes a mistake; a map of 2027–2035 along five axes, three scenarios with benchmarks, and a strategy in which the human signs the rule and MITR checks every action. Every claim is labeled: measured, being built, plan, forecast.

About the book and chapter formulas → · Architectural memorandum (PDF, 35 pages, in Russian) · Book excerpt (PDF, 60 pages, in Russian) · E-book on LitRes (in Russian) · Independent verification protocol

Public channels

Engineering and research profiles: ORCID · fishchuk.org · Habr · VC.ru · Telegram

Contact the institute: af@isslab.ai (pilot and inquiry brief composer). About the institute: legal details, subject of work, cooperation.