Vietnam AI law sets a risk-based model for AI governance

At a glance

What this is: Vietnam’s AI Law establishes the country’s first comprehensive AI governance framework, using risk tiers, role-based accountability, and extra obligations for high-risk systems.

Why it matters: For AI governance, compliance, and security teams, the law shows how regulatory expectations are moving from policy intent to lifecycle controls, evidence, and incident handling.

Context

Vietnam’s AI Law addresses a governance gap that many organisations still manage inconsistently: deciding which AI systems need stronger controls, who owns them, and what evidence proves they are safe enough to deploy. The law makes that decision structure explicit by tying obligations to risk and lifecycle role.

For practitioners, the important shift is that AI governance is no longer confined to a policy layer. When a law defines developers, providers, deployers, and users separately, it creates a control model that has to be reflected in approvals, monitoring, documentation, and incident response across the AI estate.

Key questions

Q: What breaks when high-risk AI systems are not governed with ongoing risk management?

A: Without continuous risk management, high-risk AI can produce biased, unsafe, or opaque outcomes that fail legal expectations and damage trust. The main breakdown is not just technical error, but unmanaged lifecycle drift, where data, model behavior, and deployment conditions change over time. Organisations need recurring assessments, controls, and human intervention paths to keep risk acceptable.

Q: Why does lifecycle accountability matter in AI governance?

A: Lifecycle accountability matters because AI risk is shared across developers, providers, deployers, and users. If ownership is unclear at any handoff, no one can prove who approved the system, who maintains controls, or who responds when harm occurs. That makes traceability a governance requirement, not an administrative nice-to-have.

Q: What are the signs that AI governance is failing in the enterprise?

A: Common warning signs include rapid growth in AI use without matching policy coverage, sensitive files being copied into personal accounts, and a large share of AI apps carrying high or critical risk. Another indicator is weak visibility into who is using which tools and what data they are sending. If teams cannot answer those questions, governance is not working as intended.

Q: How should organisations handle serious AI incidents under a risk-based regime?

A: They should predefine the conditions that trigger escalation, preserve the evidence needed to explain what happened, and route the case through legal, security, and compliance owners. A serious incident process is effective only when it can move from detection to reporting without reconstruction.

Technical breakdown

Risk-based AI governance and conformity assessment

Vietnam’s model classifies AI systems by risk level, then attaches stricter obligations to high-risk uses. That matters because risk-tiering turns governance into a deployment gate rather than a generic policy overlay. Conformity assessment is the mechanism that tests whether a system meets stated safety and compliance requirements before it is allowed into operation. For lower-risk uses, the law leaves more flexibility, but for higher-risk systems the organisation must prove that safeguards exist and are working. In practice, this pushes AI governance closer to assurance, where evidence matters as much as intent.

Practical implication: build risk classification and approval checkpoints into AI intake so no system bypasses pre-deployment review.

Lifecycle accountability across developers, providers, deployers, and users

The law’s role model is important because modern AI systems are rarely owned by a single team from design to production. Developers build or train the system, providers place it on the market, deployers operate it in a business context, and users consume its outputs. Each role can carry different responsibilities, which closes the common accountability gap where everyone assumes someone else owns the control. This is especially relevant where third-party AI is embedded into hiring, credit, customer service, or operational workflows. Governance has to follow the system across the supply chain and into live use.

Practical implication: map every AI system to an accountable owner, a provider relationship, and a deployer control set before production use.

Incident handling, transparency, and the governance evidence trail

High-risk AI systems are subject to ongoing compliance expectations that include logging, documentation, transparency, and incident handling. That combination matters because AI governance fails when there is no evidence trail to reconstruct what the system did, what data it used, and who was responsible when harm emerged. The law also treats serious incidents as reportable events when AI operation causes or risks significant harm. That moves AI oversight into the same operational territory as resilience and regulatory response, where detection, escalation, and documentation must already exist before an incident occurs.

Practical implication: treat AI logging, documentation, and incident workflows as mandatory control evidence, not optional technical extras.

NHI Mgmt Group analysis

Risk tiering is the right governance primitive for AI, but only if organisations can evidence it. The Vietnam AI Law aligns obligations to the likely impact of a system, which is far more workable than a one-size-fits-all rule set. That also means governance teams need reliable classification criteria, documented review decisions, and a repeatable approval path. If those records do not exist, the risk-based model becomes symbolic rather than enforceable.

Lifecycle accountability is now the core control gap that AI governance has to close. The law’s separation of developers, providers, deployers, users, and affected persons reflects how AI actually moves through enterprise environments. The governance failure it addresses is role ambiguity, where compliance breaks because no team owns the system after procurement or integration. Practitioners should treat accountability mapping as a control, not a formality.

Serious incident reporting turns AI operations into a regulatory evidence problem. When a law expects organisations to report harm, they need logging, traceability, and escalation paths that survive operational pressure. That is especially relevant for AI systems used in public service, finance, and other high-impact workflows. The practical lesson is that AI governance must be measurable at runtime, not only documented at design time.

AI governance debt will become a real programme risk for organisations that deploy third-party models without control mapping. The law makes it harder to assume that provider assurances alone are sufficient once a system is in live use. That intersects with identity and access governance because deployers still control the operational environment, the permissions, and the oversight process. Practitioners should expect more scrutiny on who can change prompts, models, data inputs, and decision thresholds.

Human oversight remains the decisive safeguard where AI affects rights, safety, or public trust. Vietnam’s framework reinforces the idea that AI should support decision-making rather than silently replace it in higher-stakes contexts. That creates a governance requirement for review points, escalation paths, and accountability when outputs are uncertain or harmful. Teams should design controls so a human can intervene before the system’s decision becomes irreversible.

From our research library:

• 52% of respondents see AI security decision-making power shifting toward platform and infrastructure teams rather than the executive suite, according to the 2026 Infrastructure Identity Survey.
• Only 44% of organisations have implemented any policies to manage their AI agents, despite 92% agreeing that governing AI agents is critical to enterprise security, according to the 2026 Infrastructure Identity Survey.
• Read next: Agentic AI Identity Risk Board Briefing

What this signals

Risk-based AI regulation will push control ownership closer to platform and infrastructure teams. Governance models that rely on executive sign-off alone will struggle, because the operational evidence now lives in deployment pipelines, model settings, and monitoring records rather than policy decks. Teams should expect AI oversight to become a shared security and engineering responsibility, not a periodic review exercise.

Lifecycle mapping is becoming a core control, not an administrative artifact. Once laws distinguish developers, providers, deployers, and users, organisations need to know exactly where accountability starts and stops for each system. That is especially important when third-party AI is embedded in business workflows and the local team still controls the runtime environment.

AI governance debt: the gap between a policy that names risk tiers and a production environment that cannot prove them. The practical consequence is that logging, assessment evidence, and escalation paths become the controls auditors and regulators will ask for first. For organisations with agents and embedded AI, governance has to be designed into operational workflows, not appended after deployment.

For practitioners

• Risk-classify every AI system before deployment Create a mandatory intake step that assigns low, medium, or high risk using business impact, public safety, and rights impact criteria.
• Map accountability across the AI lifecycle Record the developer, provider, deployer, user, and business owner for each system so governance does not stop at procurement.
• Require pre-deployment conformity evidence For high-risk systems, keep the assessment outcome, reviewer, and supporting documentation attached to the release record.
• Build incident reporting into AI operations Define the triggers, evidence capture, and escalation path for serious incidents so reporting can happen without reconstructing facts after the event.
• Preserve logging and transparency evidence Ensure model inputs, outputs, overrides, and human interventions are retained long enough to support compliance and dispute handling.

Key takeaways

• Vietnam’s AI Law moves AI oversight from general policy into a risk-tiered control model with explicit obligations for high-impact systems.
• The law makes accountability traceable across the AI lifecycle, which closes a common gap in multi-party AI delivery and deployment.
• For practitioners, the practical task is to tie risk classification, evidence capture, and incident handling to every AI system before it reaches production.

Key terms

• Risk-Based AI Regulation: A regulatory approach that applies stricter obligations to AI systems based on their potential harm, not just their technical design. It typically increases requirements for assessments, documentation, oversight, and monitoring in high-impact use cases such as employment, finance, and essential services.
• Conformity assessment: A conformity assessment is the formal process used to show that a high-risk AI system meets the obligations required before it is placed on the market. It combines documentation review, technical verification, and evidence of operational controls, rather than relying on policy statements alone.
• AI Lifecycle: The AI lifecycle is the end-to-end path from problem framing to retirement. It covers the decisions that shape a system’s purpose, data, behaviour, deployment, oversight, and decommissioning. In practice, it is the governance map that shows where risk enters and where accountability must stay active.
• Serious AI Incident: An event linked to AI operation that causes or could cause significant harm to people, rights, property, cybersecurity, public order, or national security. It is the kind of outcome that turns AI operations into a reporting and response issue.

Deepen your knowledge

NHI Mgmt Group’s NHI Foundation Level course covers NHI governance, agentic AI identity, machine identity security, IAM, and secrets management through the industry’s only accredited NHI security programme. It helps practitioners connect identity controls to broader security and governance programmes.