Vietnam Now Has Rules for AI: What It Means for a Small Team
This publication is publicly available with an advisory disclaimer but has not yet received recorded CLO approval. Verify current law and procedure before relying on it.
General information only, current at the time of writing. Rules for foreigners and small businesses in Vietnam change - please confirm your own situation before acting on it.
Direct answer
Vietnam now has dedicated rules for artificial intelligence, and they sort systems according to the harm each could cause rather than applying one standard to everything. For a small team using ordinary tools to draft, translate and answer common questions, the practical effect is light - but it is no longer nothing.
This article stays in plain language and avoids citing provisions. The aim is for a business to recognise where its own use sits and what follows from that.
Risk determines the obligations
The framework sorts systems into tiers according to the harm they could cause. Higher potential harm brings heavier expectations around oversight, documentation and accuracy. Lower potential harm brings lighter ones.
The orientation is developmental. The intent is to encourage adoption while managing the genuinely risky end, rather than to discourage use across the board.
For a small business the practical consequence is reassuring but not zero. Most ordinary use sits at the light end, and the obligations that do apply are broadly the practices a careful operator would follow anyway. The change is that they are now expectations rather than good habits.
Where ordinary small business use sits
Drafting a customer reply. Translating a product description. Summarising a long message thread. Answering frequently asked questions from the business's own published material.
These are low-stakes applications. A system assists a person with a task and a person remains in the loop, which is the profile the lightest tier is designed around.
What keeps a business in that tier is the person in the loop. The position changes when a tool begins deciding something independently that materially affects a customer - not because the tool became more capable, but because the review step was removed. That removal usually happens for practical reasons, when volume makes checking every output feel inefficient.
"The question is not how capable the tool is. It is whether anyone reads the output before it affects somebody. That single fact moves a business between categories." - LotusGuard Vietnam Advisory
What counts as higher risk
Applications that materially affect a person's interests without meaningful human review.
| Lower risk | Higher risk |
|---|---|
| Drafting a message a person sends | Sending automated decisions with no review |
| Translating the business's own content | Deciding something about a person's money |
| Answering questions from published prices | Screening job applicants automatically |
| Summarising for internal use | Deciding who receives a service or a refund |
| Suggesting, with a person deciding | Deciding, with nobody checking |
The right-hand column is uncommon in micro-businesses, which is genuinely good news. It is not unheard of, and it tends to arrive quietly - typically when someone automates a step that previously involved a person because the volume had become tedious. That is the moment at which a business changes category, usually without anyone noticing.
Five things a small team should do
- Know which tools are in use and what data each can reach. Most teams cannot currently answer this.
- Tell people when they are dealing with an automated system, particularly in customer-facing chat.
- Keep a person in decisions that matter - money, employment, access to a service.
- Do not enter personal data into tools that have not been checked. Remove identifying details by default.
- Record what was decided. A single page listing which tools are used, what they can reach, and what they must not.
Point five carries more weight than it appears to. Across frameworks of this kind, being able to demonstrate that the question was considered forms a substantial part of the expectation, and for a small business a page is sufficient.
The page also has a practical use beyond compliance. A business that can state which tools it uses and what each touches can answer a customer's question, brief a new employee, and evaluate a proposed tool without starting from nothing.
How this relates to data rules
They are separate and both can apply at once. One set governs how personal information is handled. The other governs the automated systems themselves.
A customer chatbot connected to a customer list falls under both simultaneously: it processes personal data, and it is an automated system interacting with people. That overlap is why deciding what a tool may access is the single most useful control available to a small business - one decision addresses both.
The work is best done once. A written note of what data the business holds and which tools touch it satisfies the practical core of both sets of expectations, which is the argument for treating them as one task rather than two.
What does not warrant concern
Using a mainstream tool to assist with writing. Translating the business's own material. Requesting a summary of something already held.
None of this is what the framework addresses, and a small business using these tools carefully to reduce typing is not the risk the rules were designed to manage.
Two cautions remain. Rules in this area are new and will develop, so this should be treated as a description of direction rather than a settled position. And where a use involves decisions that meaningfully affect people, or sensitive categories of information, that is the point at which advice specific to the situation is worth more than a general article.
Frequently asked questions
Does Vietnam regulate artificial intelligence now? Yes. Vietnam has introduced dedicated rules for artificial intelligence, designed to encourage development while managing risk. The approach sorts systems according to how much harm they could cause, rather than applying one standard to everything.
Does this apply to a small business using ordinary tools? Most everyday small business uses - drafting, translating, answering common questions - sit in the lightest category and carry correspondingly light expectations. The heavier obligations are aimed at systems capable of causing significant harm.
What should a small team actually do? Know which tools you use and what they touch, tell people when they are dealing with an automated system, keep a human involved in decisions that affect someone materially, and do not feed personal data into tools you have not checked.
Is this separate from data protection rules? Yes, and both can apply at once. One set concerns how you handle personal information; the other concerns the automated systems themselves. A chatbot using customer data sits under both, which is why deciding what it may see matters.
What changes if my use is higher risk? Expectations rise with the potential for harm - more documentation, more human oversight, more care about accuracy. If your tool materially affects someone's money, employment or access to a service, treat it as higher risk and get advice.
Conclusion
The framework scales with potential harm, which places most small business use at the light end. What moves a business between tiers is whether a person still reviews output before it affects someone, and that step is usually removed for convenience rather than by decision. And a single written page - which tools, what they access, what they must not - covers most of what a small team is realistically expected to hold.
For a view on which tier a particular use falls into, talk to LotusGuard. For most small businesses the answer is reassuring, and it is worth hearing from someone with no software to sell.
