IS 3.0 glossary

Terms of the security of actions of agents built on artificial intelligence (AI), as they are used in the book Proven Action (in Russian) and in the section of this site on information security of the third generation (IS 3.0), followed by the terms of the method of the institute. Chapter and appendix references are to the book.

Terms of IS 3.0 and the book

  • MITR: a deterministic control layer for the admissibility of an agent’s actions. It checks every tool call against contracts written as data, using a solver, and writes a receipt to a hash chain (chapters 8–12). See MITR.
  • INDRA (Invariant, Nondestructive, Deterministic Resilience Audit; page in Russian): a bench that tests web services and application programming interfaces with safe probes and turns a vulnerability it finds into a draft rule for MITR (chapters 13–14). Its acceptance on 25 October 2026 met 12 of 12 criteria on an internal reference bench of 28 labeled scenarios, prior to external accredited laboratory validation.
  • The “thought to action” watershed: the boundary at which what is checked is not the model’s reasoning but the finite set of parameters of a call to an external tool.
  • Contract as data: a rule written declaratively in a data file rather than in program code; as of 24 October 2026 MITR holds 20 contracts (19 on 14 October 2026).
  • Satisfiability solver (SMT, satisfiability modulo theories): a program that checks the satisfiability of logical formulas; in MITR it is Z3 (developed by Microsoft Research), and it runs on an ordinary processor. See formal verification.
  • unsat: the verdict “unsatisfiable”: the intersection of the observed facts of the action with the set of forbidden states is empty, so the action is admitted.
  • sat: the verdict “satisfiable”: a counterexample of a contract violation has been found, so the action is blocked, and the agent receives the violated rule and the conditions for a fix.
  • Advisory mode: the solver checks every call and writes a receipt, but the agent is not blocked, and statistics of false blocks build up. The mode is set for each class of actions, and since 24 October 2026 the default for all classes is enforce mode.
  • Enforce mode: a call with the sat verdict is not executed, and a class of actions is moved to this mode when the share of false blocks falls below a threshold. As of 24 October 2026, all 20 contracts run in enforce mode.
  • Receipt: a record of a decision, both “yes” and “no”, in the chain with an external timestamp, which can be verified without trusting the owner of the server.
  • Hash chain: a sequence of records in which each one contains the SHA-256 hash of the previous one (SHA stands for Secure Hash Algorithm), so editing any record breaks all the following ones.
  • OpenTimestamps anchor: an external timestamp for the state of the chain: after it, the owner of the server cannot quietly rewrite the history. In regulated Russian banking and critical infrastructure environments, OpenTimestamps is paired with accredited trust service time stamps and qualified electronic signatures under Federal Law No. 63-FZ.
  • Tenant: a client of MITR with its own access token and an isolated loop.
  • “Not measured”: the third value of admission (⊥) next to “yes” and “no”: the absence of a signal is not counted as “clean” and is not executed (chapter 11).
  • Truth marking labels: “measured”, “building”, “concept” and “forecast”, and for monetary claims in Appendix A of the book, “fact” and “hypothesis”.
  • Branches A, B and C: forecast scenarios, where A is “normal technology”, B is “continuation of the trend”, taken in the book as the base branch, and C is “acceleration”.
  • Benchmark: the date and source by which it is checked whether a forecast came true (Appendix D of the book).
  • Gate (of a stage): an acceptance criterion for a stage with a deadline, counted by an instrument, for example the acceptance gates of the INDRA bench.
  • “Six out of six”: six basic claims about MITR that a daily instrument confirms or does not confirm. All six held for the first time for 23 October 2026 (on 9 October 2026, one of six), and for 24 October 2026, five of six were confirmed, with one not measured.
  • Verifier for an outsider: a single-file program that recomputes the chain and replays the decisions of MITR with any solver that understands the SMT-LIB format, without a line of MITR code. In the historical epoch-33 anchor on 25 October 2026, 5,079 of 5,079 decisions matched (on 14 October 2026, 4,958 of 4,958); as of 15 November 2026, the live chain holds 5,373 records across 55 completed epochs.
  • Viable system model (VSM) of Stafford Beer: System 1 (S1) handles operations, System 2 (S2) coordination, System 3 (S3) management, the S3* channel independent audit and admission, System 4 (S4) intelligence, and System 5 (S5) policy.
  • S3*: in the viable system model, the channel of audit and admission independent of operational management; this is the place of MITR.
  • Algedonic channel: an emergency pain signal that bypasses the hierarchy; in MITR it corresponds to the rule “unmeasured means unexecuted” with the decision passed to a human (S5).
  • Levels of authority: the Chinese three-level scheme (decisions only by the user, on the user’s instruction, by the agent on its own; see China’s rules for AI agents); in MITR, the advisory and enforce modes with a move between them by the share of false blocks.
  • Guardian agent: the term of Gartner for systems that watch, check and limit other agents; in these terms MITR is a deterministic guardian over contracts as data.
  • Agent identification (KYA): establishing the identity of an agent and the authority delegated to it, by analogy with the customer checks of banks; in MITR it corresponds to the tenant and its access token.
  • Network jurisdiction: a community with its own machine-readable law and control layer, recognized by an agreement with the state (a forecast).
  • MCP, A2A, AP2, x402: protocols of the 2026 internet of agents that provide connection of tools (MCP), interaction between agents (A2A) and agent payments (AP2, x402).

Terms of the institute’s method

The terms the institute uses in its Doctrine, in the description of its method (both in Russian) and in its work. For each term, the sense in which it is used and the tradition it comes from are given, and terms that have a precise meaning in mathematics are used either in that meaning or with an explicit reservation.

System synthesis

A method of assembling systems from conflicting requirements in which the conflict is removed by separating the requirements along different coordinates of the system rather than by averaging. It applies only when there is a structural contradiction, and it is not a synonym for “integration” or “hybrid”: integration joins what is compatible, while synthesis joins what is incompatible, at the price of introducing an additional coordinate.

Invariant

A property that a system must hold in its external environment regardless of changes in its composition and state. It is used here in a sense close to a loop invariant in the theory of programs: a statement that is true before and after each step. An invariant describes an obligation of the system independently of its internal design or intention.

Structural contradiction

A pair of requirements in which one requires a property to be present and the other requires it to be absent. It goes back to the technical and physical contradiction of the theory of inventive problem solving (TRIZ). It differs from a conflict of interest: a conflict of interest belongs to the parties, while a structural contradiction belongs to the system itself and persists with any set of participants.

The term “topological rupture”, which appears in working materials as an image, is not used on these pages: topology has a precise mathematical meaning that does not apply to the phenomenon described.

Separation coordinate

The dimension along which conflicting requirements are met separately: the level of the system, the phase of time or the control loop. A direct application of the independence axiom of axiomatic design, in which functional requirements are distributed over independent design parameters.

Feedback loop and its closure

A loop is considered closed if an error at the output of a block becomes a corrective signal at the input of the preceding block without an external operator. A loop that needs a human to pass the signal on or to make the correction is considered open, and it is counted as such when the system is assessed. There are no intermediate states in this assessment.

Algedonic channel

The channel of an emergency signal in Stafford Beer’s viable system model: the path by which a signal of pain or threat passes through the levels of management, bypassing the usual approval procedure. It is used here in its original meaning. The presence of such a channel does not make a loop closed if the recipient of the signal is still a human.

Requisite variety

William Ross Ashby’s law: the variety of the controlling loop cannot be less than the variety of the controlled object, otherwise control is impossible. It is used in its original meaning and acts as a constraint when choosing a separation coordinate: moving a requirement to a level whose variety is insufficient leads to a loss of controllability.

Ideal final result

The form of stating a goal in the theory of inventive problem solving: the function is performed, but there is no system performing it. It is used as a guide for formulating an invariant rather than as an attainable state.

Deterministic verification

A check of the admissibility of an action that gives the same result on the same input data and rests on a mathematical proof rather than a probability estimate. It is opposed to probabilistic filtering, where the same input can lead to different decisions. A refusal by such a check blocks the action before it is executed, rather than signaling afterwards.

Nearly decomposable system

Herbert Simon’s notion: a system in which the links within a level are substantially stronger than the links between levels. It provides the basis for separation by level, so if the links between the supposed levels are comparable to the links within them, the levels have been identified wrongly, and the separation will be illusory.

Emergence

The appearance in the assembled whole of properties that none of its parts has. It is used in a descriptive sense, without strong philosophical assumptions. In organizations, unlike in materials, emergence is not checked by testing a sample: only the consequences can be checked, and only with a delay.

The full account of the procedure, the conditions of refusal and the limits is on the page “The method of system synthesis” (in Russian).