Quang NinhThe digital transformation management agency of Vietnam and Australia discussed the promotion and governance of AI in the two countries, and shared experiences, including high-risk AI governance.

The issue was raised at the Vietnam - Australia Digital Cooperation Forum, within the framework of Vietnam International Digital Week 2026, held in Quang Ninh on October 28.

In the discussion, Mr. Tom Marwick from the Australian Digital Transformation Agency asked how Vietnam combines an AI classification framework by risk level with a real-world impact assessment tool, and how to coordinate to both ensure common standards and create flexibility in the implementation process.

"This is the complex problem that Australia is also facing: how can a diverse administrative system operate synchronously, with clear standards to simplify the process, but still have enough flexibility to effectively manage risks arising right from the grassroots," he said.

Sharing this issue, Chu Thi Tham, from the National Institute of Digital Technology and Digital Transformation, said that Vietnam determines high-risk AI based on its ability to have a significant impact on health, human life, legitimate rights and interests of organizations and individuals, public interests and national security.

In Vietnam, the issued high-risk AI list includes 46 systems. Some systems must be assessed for suitability by an independent third-party testing organization before operation; The rest can be evaluated by the supplier themselves or hired by a third party before being released to the market.

Based on common criteria, specialized management ministries propose a list of high-risk AI systems according to the specific fields in charge. The Ministry of Science and Technology evaluates proposals before finalizing and submitting them to the Government for promulgation.

She also emphasized that AI governance is not intended to prohibit or inhibit technology, but to create an environment that promotes application, while controlling negative impacts. Risks do not just lie in the algorithm but can also stem from unrepresentative training data, biases in the data, fake content generated by AI, invasion of privacy, or attacks aimed at distorting data and models.

Regarding the public sector, Ms. Tham affirmed the principle that people must retain the right to make decisions and bear the final legal responsibility. State agencies need to assess the impact on high-risk AI systems that have a major impact on social justice or public interest, while ensuring transparency, processing and protection of public data.

Sharing more, Mr. Hoang Huu Hanh, Deputy Director of the National Digital Transformation Department, said that the high-risk AI list cannot be a fixed list but needs to be reviewed and updated according to the development of technology. A system may initially be classified as low risk when directly controlled by humans, but the level of risk may change as automation increases, human intervention decreases, and the system begins to impact life safety or national security.

He also mentioned that the controlled testing mechanism can be applied as a tool to verify new systems without clear risk boundaries before being put into large-scale management.

Vietnam - Australia share experiences in AI transformation

At the forum, representatives of both sides emphasized the need to make the governance mechanism a part of the technology application process, instead of only handling it when risks have occurred.

Speaking at the event, Ms. Lucy Poole, representative of the Australian Digital Transformation Agency (DTA), raised two key issues: building a digital government capable of operating effectively in the long term and applying AI that is developing rapidly in a responsible way, with an accountability mechanism established from the beginning.

Australia currently uses an AI governance toolkit consisting of three components. In particular, AI technical standards provide requirements throughout the entire life cycle of the system; AI Impact Assessment tool helps agencies identify risks in each use case; The AI ​​Review Panel advises on high-risk, novel or complex applications.

The DTA representative also said that Australia's AI impact assessment tool is built on eight AI ethical principles, which are mandatory requirements in the government's AI policy for all use cases. Accordingly, although the initial assessment step may be subjective on the part of the recommending agency, this tool provides an analytical framework to identify potential risks. Regular use also helps staff improve their self-assessment capacity.

Meanwhile, the AI ​​Review Council does not act as a sanctioning body but plays an expert advisory role. The council's recommendations are not legally binding, but agencies often proactively adjust projects according to expert opinion before implementation.

Mr. Hoang Huu Hanh said that Vietnam has shaped the legal framework and AI management orientations based on the level of risk. The next challenge is to put the regulations into practice, which requires resources and implementation experience. Vietnam hopes to continue to exchange with partners such as Australia to improve enforcement capacity.

Liu Quy