Vietnam Mandates Local LLMs in New Public Sector AI Governance Framework

Vietnam's Minister of Science and Technology has issued Decision No. 2847/QD-BKHCN, introducing a detailed evaluation framework for national-level Artificial Intelligence platforms used in public services. The new criteria will serve as a benchmark for AI integration across 14 ministries, three ministerial-level agencies, and 34 local governments.

The framework utilizes a 100-point scoring system split into two sections. General criteria, designated as Group A and worth 35 points, cover legal compliance, source citation, data boundaries, AI security, personal data protection, and ethics. Group B, accounting for the remaining 65 points, assesses specialized functions, including the platform's capacity to analyze and summarize official government documents and sector-specific regulations.

To gain approval for official deployment, a platform must score a minimum of 70 points, secure at least 50% of the available points in Group A, and have no disqualifying marks. Systems scoring between 50 and 70 points may enter a trial period but must be upgraded and re-evaluated within two months. Platforms scoring below 50 points will be considered ineligible for testing.

A key provision in the new rules emphasizes technological autonomy. To qualify for even a trial, platforms must be built on Vietnamese Large Language Models (LLMs) that are developed and owned by domestic companies. The decision also mandates that all related infrastructure for training, inference, and service delivery must be physically located within Vietnam.

Beyond technical specifications, the ministry also requires that platforms feature robust protocols for reporting serious incidents. Performance standards are stringent: systems must support at least 5,000 concurrent users during the trial phase and scale to a minimum of 50,000 concurrent users for full implementation.