Deputy Minister of Science and Technology Trần Trung Tính (middle) chairs the ministry's monthly press briefing in Hà Nội on Wednesday. — Photo courtesy of the ministry

Bảo Hoa & Khánh Dương

HÀ NỘI — Việt Nam's Ministry of Science and Technology has responded to AI developers and the European Commission's proposal to 'pace the frontier', stressing that Việt Nam has established a sufficient AI regulation framework that emphasises human control.

In response to Việt Nam News reporters' question about the call from OpenAI and Anthropic to slow down AI development and the State of the Union address by EC President Ursula von der Leyen, the ministry on Wednesday said Việt Nam's approach to AI is similar to that of the EC, which is to continue to 'harness the benefits of the technology, while simultaneously enhancing safety and governance capabilities commensurate with the risks'.

"It is essential to properly assess recent international developments," said the National Institute for Digital Technologies and Digital Transformation (NIDiT), the ministry's research body, in a written response released after the September press briefing.

"While advanced artificial intelligence systems present significant new risks, it can't be concluded from a specific incident that artificial intelligence in general has spiraled out of human control," the institute said.

The institute was referring to the cybersecurity incident disclosed by OpenAI in July in which its AI agents, while being tested in an internal cyber-capability evaluation, escaped their controlled testing environment and hacked into the infrastructure HuggingFace, a collaborative coding platform dedicated to the AI developers community.

Following a warning from a former Anthropic and OpenAI researcher about the worrying development speed of 'self-improving superintelligence', Anthropic's chief executive officer Dario Amodei on September 12 called to 'slow the pace' of AI model improvement, but also adding that it does not mean 'halting model training or technical progress' but giving companies time to safeguard their models, certified by third-party evaluations.

OpenAI took a similar approach, announcing policy to enhance independent evaluations of its AI model training, assessment and deployment processes on September 22.

In her address to the European Union on September 16, EC President von der Leyen proposed to team up with 'like-minded partners' like Canada and the United Kingdom on AI model evaluation, verification, early warning and security, while reitarating that Europe's stance is 'to make the most of AI' to improve society, quality of life and productivity.

"The common thread among these messages is the continued harness the benefits of the technology, while simultaneously enhancing safety and governance capabilities commensurate with the risks," said NIDiT, adding that Việt Nam has taken similar approach with the Law on Artificial Intelligence and the National Artificial Intelligence Ethics Framework, which both took effect in March.

The AI law identifies AI as a key driver of growth while encouraging controlled testing and placing humans at the center.

It also stresses that AI does not replace human authority or responsibility, and that humans must maintain the power to oversee and intervene in the decisions and actions taken by AI systems.

"In short, while a system may become increasingly technically autonomous, legal liability does not shift to the machine," the institute said.

"Developers, providers, deployers and users must still fulfill their respective obligations, and authorised individuals remain accountable for decisions made within their authority."

In addition to legal obligations, the national AI ethics framework provides guidance on safety-by-design, pre-deployment testing, data protection, human oversight and response strategies for system deviations or misuse.

The framework addresses risks such as hijacking, data poisoning, model poisoning, adversarial attacks and vulnerability exploitation.

"Put this framework into practice is the current top priority," said NIDiT.

The institute proposed to closely monitor real-world evidence of tasks completed by AI agents capable of executing long task sequences and utilising tools, for timely assessment and management.

"Our stance is to foster the development and application of artificial intelligence while ensuring that human capabilities for assessment, safety assurance and control evolve in tandem with technological capabilities," it said.

Low-risk applications should be allowed, whereas high-risk systems must undergo commensurate assessment and monitoring, it added.

"In this way, Việt Nam can protect its people while simultaneously seizing the opportunities for innovation and growth offered by artificial intelligence." — VNS