Five Russian GOST Standards for AI Analyzed
Since January 1, 2025, Russia has enacted a package of national standards for artificial intelligence, all adapted from international ISO/IEC documents. The five standards cover AI management systems, lifecycle processes, knowledge engineering architecture, and data quality (parts 1 and 3). While voluntary in principle, they become mandatory for government procurement and may be used as benchmarks in lawsuits.
Since January 1, 2025, Russia has implemented five national standards on AI, all developed by Technical Committee TC 164 and approved by Rosstandart on October 28, 2024. The standards are adaptations of international ISO/IEC documents: one is identical (IDT), four are modified (MOD). GOST R ISO/IEC 42001-2024 (AI management system) is the only IDT standard; its appendices specify nine control domains including AI policies, resource management, impact assessment, and lifecycle safety (data poisoning, model theft). GOST R 71539-2024 (AI system lifecycle processes) breaks down 17 technical processes from mission analysis to disposal, with special attention to data engineering (different test/train sources), continuous validation (automatic rollback for continuous learning), and decommissioning data handling. GOST R 71540-2024 (knowledge engineering reference architecture) addresses ontologies, knowledge graphs, and symbolic AI, with six aspects (safety, reliability, etc.) and a nine-process knowledge chain. The GOST R 71484 series covers data quality: Part 1 defines a six-stage data lifecycle model with requirements for bias mitigation and data security; Part 3 mandates data quality culture, competence management, audit planning, and minimization principle. The standards are voluntary by law but become mandatory if referenced in government contracts (44-FZ, 223-FZ) or used as expert references in court.
- Сокращения
- IDT = Identical — Идентичный оригиналу
- MOD = Modified — Модифицированный
- ISO = International Organization for Standardization — Международная организация по стандартизации
- IEC = International Electrotechnical Commission — Международная электротехническая комиссия
- GPU = Graphics Processing Unit — Графический процессор
- MLOps = Machine Learning Operations — Операции машинного обучения
- LLM = Large Language Model — Большая языковая модель
- RAG = Retrieval-Augmented Generation — Генерация с дополнением извлечением
Source: Habr — хаб ИИ —
original