FSTEC Order No. 117 and artificial intelligence

Jurisdiction Case Study · High-Regulated Infrastructure & Comparative AI Law

Order No. 117 of 11 April 2025 of the Federal Service for Technical and Export Control of Russia (FSTEC of Russia), the regulator for the technical protection of information, has applied since 1 March 2026 to state information systems and other systems of state bodies, state unitary enterprises and state institutions. Its requirements on artificial intelligence (AI) are set out in points 60 and 61. The draft amendments of 24 August 2026 restate these points and introduce measures for AI models and agents from 1 March 2027. Below is what the text of the draft says and how its requirements can be met.

Quotations from the order and the draft below are a working translation by the institute; the official texts are in Russian.

This page covers specifically points 60 and 61 of Order No. 117 (in force since 1 March 2026) and the 24 August 2026 draft amendments (#170500) governing artificial intelligence models and autonomous agents. Pre-execution mathematical admission is detailed on the preventive gateway page, and isolated adversarial bench testing is documented on the “Indra” verification stand.

What the draft amendments to Order No. 117 require

The draft was published on the federal portal of draft regulations under No. 170500 (in Russian) on 24 August 2026 and is to take effect on 1 March 2027. It had not been approved as of 16 October 2026. Checked: 16 October 2026, against the text of the draft.

  • Point 14(d): the documents of the operator must include the procedure for protecting information when artificial intelligence is used.
  • Point 60: AI models are placed in a separate segment, privileged access to it requires strengthened multi-factor authentication, users work with models with minimal rights, and their actions in the segment are logged and analyzed.
  • Point 60, for models that process restricted information or perform significant functions of the operator: these measures are joined by control of access to models, filtering of requests and responses, request quotas, and limitation and control of the functionality of the models.
  • Point 61: “when AI agents are used, including autonomous AI agents”, the operator takes measures against unauthorized impact on the resources of the information system and controls and manages the access rights of the agents under internal standards and regulations.
  • Definitions: artificial intelligence is understood as in Article 3, point 1 of Federal Law No. 243-FZ of 26 July 2026 (see Russia’s law on artificial intelligence), and an AI agent as in clause 3.1.1 of the Russian national standard (GOST) R 71476-2024.

What an operator lacks

A typical situation at an operator of a state information system, whether a regional agency, a state institution or a state unitary enterprise, is as follows. An agent on a local model has already been deployed, but the security team has no tool that would keep a log of the calls of the agent that cannot be quietly edited in places and would show that the rights of the agent are limited and each of its actions is checked. An ordinary system log can be edited with one command, and no trace of the edit remains.

How the requirement is met

The open log iss-ledger-117. Receipts of actions form a hash chain under GOST R 34.11-2012 (“Streebog”), and an outsider recomputes its integrity with a single-file verifier in an isolated environment, without trusting the developer and without installing the software of the developer. The package, the source code and the verifier are below.

Open source · Apache-2.0 · iss-ledger-117 1.1.1

Installation package (.whl)

iss_ledger_117-1.1.1-py3-none-any.whl · sha256: 8d4ffe60f41ebd377c476f369655aaef5e0ac91858a3d294be37fd1e3100bc87

Source code (.tar.gz)

iss_ledger_117-1.1.1.tar.gz · sha256: 6eccef523e24fa60fed29c78866ff12afbab092bd17d51642e00d34dfcea9fa5

Single-file verifier

iss_verifier.py · sha256: 1ff894574407307d33ec99115652bcaec4e53d06189178696d0db1c767a8608f

Source code: https://git.isslab.ru/isslab/iss-ledger-117

Measurement: the implementation passes the test examples from the text of the standard, and the single-file verifier works in an isolated environment without a network.

The open package mcp-sentinel sits between the agent and the tool server and does not let a call through to execution if it matches a prohibition pattern: a destructive database query, a step outside the working folder, a leak of a secret or a dangerous shell command. This is pattern matching, not a solver, so a call that has been changed beyond recognition may get through, as the description of the package states. The full version of MITR, in which the admissibility of an action is decided by a solver under a contract, is provided for a pilot.

Open source · Apache-2.0 · mcp-sentinel 1.1.2

Installation package (.whl)

mcp_sentinel-1.1.2-py3-none-any.whl · sha256: 681a4db59a1a44e89a2b1386ec4b4f2e1fc23c1025c295b9ebf00fd5d20901ea

Source code (.tar.gz)

mcp_sentinel-1.1.2.tar.gz · sha256: abb88978db7b34a2f1f56e0eff42bcd1365b9cc0eb17e4a4380f7815e4f1b79e

Source code: https://git.isslab.ru/isslab/mcp-sentinel

Measurement of 21 September 2026 on 100 calls: checking contracts in memory takes 0.015 milliseconds at the median, and a full cycle including writing the decision to the log on disk takes 22.5 milliseconds.

Scope of the open packages: The open packages help keep and verify a log of agent actions, but they do not certify compliance with the order, which is established by the organization together with the certification body. The hash function implementation passes the test examples of GOST R 34.11-2012 (RFC 6986), but it is not a certified cryptographic information protection tool.

Order No. 117 and Draft #170500 requirements matrix

The table maps current Order No. 117 clauses (in force since 1 March 2026) and Draft #170500 amendments (scheduled for 1 March 2027) to deterministic software controls and explicit verification boundaries.

Order clause & statusOperator obligationTechnical closure moduleBoundary and timeline
Point 14(d)
Draft #170500
Establish internal information-protection procedures for artificial intelligence usageDeclarative MITR gate contracts separated from model weights and segment access rulesOperator policy update; draft target date 1 March 2027
Point 60 (in force)
Since 1 March 2026
Prevent disclosure of restricted data to model vendors and block tampering with data or modelsOn-premises deployment of MITR and local models inside the customer network perimeterEnforceable statutory requirement since 1 March 2026
Point 60 (restated)
Draft #170500
Isolate models, enforce multi-factor authentication, minimal rights, logging, request filtering, and quotasHash-linked iss-ledger-117 under GOST R 34.11-2012 (RFC 6986) and MITR solver invariantsSeparate certification applies if regulated cryptography is required
Point 61 (in force & draft)
GOST R 71476-2024
Validate requests, detect unreliable outputs, and govern autonomous agent access permissionsOpen mcp-sentinel filter (0.015 ms) or full MITR solver (1.08 ms median across 20 rules)Pattern package covers explicit checks; full invariants require MITR solver

Legal reference: state systems, critical information infrastructure and open code

Order No. 117 was approved on 11 April 2025, registered by the Ministry of Justice on 16 June 2025 and has been in force since 1 March 2026 in place of Order No. 17. It sets requirements for protecting information in state information systems and other systems of state bodies, state unitary enterprises and state institutions, including systems with AI (points 60 and 61).

For significant objects of critical information infrastructure, the relevant acts are different: Federal Law No. 187-FZ of 26 July 2017 “On the Security of the Critical Information Infrastructure of the Russian Federation” and Order No. 239 of FSTEC of Russia of 25 December 2017. As of 2026, no separate approved methodology for certifying software with AI technologies has been adopted.

  • One’s own development environment. No license for information protection tools is required to protect one’s own development environment; the decision is made by the head of information security under the internal policy of the organization.
  • Production use at significant objects. Here the existing procedures apply: conformity assessment of the software under the general requirements and certification of the information system that includes it. Open code and an external integrity verifier make it easier to confirm that the solution can be trusted, because the inspector does not have to rely on the word of the supplier.

This is a reference to the norms in force, not a legal opinion; the organization decides whether it applies to a specific object.

Frequently asked questions

Is a license from FSTEC of Russia needed to keep such a log under Order No. 117?

No license for information protection tools is required to protect one’s own development environment, and the head of information security decides under the internal policy of the organization. The final legal conclusion belongs to the lawyer of the organization.

Does the log work in an isolated environment without a network?

Verification runs completely offline: the package is written in Python with no external dependencies, and the verifier is a single file that can be moved to an isolated machine and executed without network access. It needs only a Python 3 interpreter, so it runs on any operating system where such an interpreter is installed.

What do the draft amendments require for AI agents?

Point 61 of the draft requires, when AI agents are used, including autonomous ones, measures to prevent unauthorized impact on the resources of the information system and control and management of the access rights of the agents under internal standards and regulations. The draft is to take effect on 1 March 2027 and had not been approved as of 16 October 2026.

How does the open mcp-sentinel package differ from the full pre-execution gateway?

The open mcp-sentinel package matches call arguments against four static prohibition patterns in 0.015 milliseconds without formal constraint solving, whereas the full MITR gateway evaluates twenty active security contracts in the Z3 theorem prover in 1.08 milliseconds at the median and seals every decision into a verifiable cryptographic receipt chain (accuracy and benchmark corpus).

Sources

  1. Draft amendments to Order No. 117 of FSTEC of Russia, federal portal of draft regulations, project No. 170500 (in Russian): https://regulation.gov.ru/projects/170500
  2. Order No. 117 of FSTEC of Russia of 11 April 2025 (in Russian): http://publication.pravo.gov.ru/document/0001202506170011

Checked: 16 October 2026