|
THE PRIME
MINISTER OF VIETNAM
-------
|
THE SOCIALIST REPUBLIC OF VIETNAM
Independence – Freedom – Happiness
---------------
|
|
No: 33/2026/QD-TTg
|
Hanoi, June 30, 2026
|
DECISION
ON ISSUANCE OF THE LIST OF HIGH-RISK
ARTIFICIAL INTELLIGENCE SYSTEMS
Pursuant to Law on
Government Organization No. 63/2025/QH15;
Pursuant to the Law on
Artificial Intelligence No. 134/2025/QH15;
Based on Government
Decree No. 142/2026/ND-CP on elaboration and measures for the implementation of
the Law on Artificial Intelligence;
At the request of the
Minister of Science and Technology;
The Prime Minister
hereby issues a Decision on the List of high-risk artificial intelligence
systems.
Article
1. Scope
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
2. Artificial intelligence
systems must satisfy the basic operating principles specified in laws on
artificial intelligence; the use of artificial intelligence systems must not
alter, transfer, or exclude the authority and responsibility of competent
authorities, organizations, or individuals as prescribed by law; ensure human
supervision, control, and intervention during system operation; and not engage
in any prohibited acts as prescribed by law.
Article
2. Criteria for determining high-risk artificial intelligence systems
Artificial intelligence
systems shall be determined to be high-risk based on the following criteria:
1. The system may cause significant
harm to the life, health, and lawful rights and interests of organizations and
individuals, national interests, public interests, or national security as
prescribed in Article 9 of the Law on Artificial Intelligence and Clause 1 of
Article 8 of Decree No. 142/2026/ND-CP on elaboration and measures for the
implementation of the Law on Artificial Intelligence.
2. The system is not
excluded under Clause 2 of Article 8 of Decree No. 142/2026/ND-CP.
Article
3. Effect and responsibility for implementation
1. This Decision comes
into force from August 15, 2026.
2. Ministers, Heads of
ministerial agencies, Chairpersons of People's Committees of provinces/cities,
and heads of relevant agencies, organizations, and individuals shall be
responsible for implementing this Decision.
Article
4. Transitional provisions
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
a) Before September 1,
2027, for artificial intelligence systems in the fields of healthcare,
education, and finance;
b) Before March 1, 2027,
for artificial intelligence systems in the remaining fields within the List.
2. During the transitional
period prescribed in Clause 1 of this Article, the aforementioned artificial
intelligence systems shall be permitted to continue operating, except where the
competent state authority determines that the system poses a risk of causing
serious harm and requests the suspension or termination of its operation in
accordance with Clause 2 of Article 35 of the Law on Artificial Intelligence.
3. For artificial
intelligence systems put into operation within 06 months from the effective
date of this Decision, the provider and the deployer must fulfill their
compliance obligations under laws on artificial intelligence before March 1,
2027.
PP. PRIME MINISTER
DEPUTY PRIME MINISTER
Ho Quoc Dung
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
LIST OF HIGH-RISK ARTIFICIAL
INTELLIGENCE SYSTEMS
(Attached to Decision No. 33/2026/QD-TTg dated June 30, 2026 of the Prime
Minister)
NO.
Name of the system
Description
Conformity assessment
Under Point b of Clause 2 of Article 13 of Law on
Artificial Intelligence
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
I
Education
1
AI system for providing
self-study content under an education program, using unregulated data sources
Assessments are
carried out only where the AI system:
- Automatically provides
content serving learners' self-study activities under an education program;
- Feeds the
learning content data without control;
- Is deployed at a large
scale capable of creating systemic risk or having a wide-ranging impact on
learners.
x
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
2
System using AI to
automatically test, evaluate performance, and rank learners
Assessments are
carried out only where the AI system:
- Is used to
automatically test and evaluate learners;
- Provides the results
that are used as the official grounds for evaluating learning outcomes,
assessing competency, or ranking learners;
- Provides evaluation
within the national education system.
x
3
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
Assessments are
carried out only where the AI system monitors and analyzes learner behavior:
Using biometric data
(facial recognition, eye-tracking, behavior, personal emotions) or
establishes automated, high-frequency feedback or alert mechanisms concerning
attention state, personal expressions, learning results, or learner
evaluation that cause psychological stress or infringe on learners' privacy
during the learning process.
x
II
Ethnic and Religious Affairs
1
AI system for
automatically scoring, classifying, and ranking dossiers to determine
beneficiaries and areas eligible for ethnic minority policies
Assessments are
carried out only where the AI system:
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
- Assessments shall not
be carried out when the results provided by the AI system are not used as
official results for publication.
x
2
AI system for
automatically authenticating and deciding the validity of information
relating to ethnic and religious affairs
Assessments are
carried out only where the AI system:
- Automatically checks,
cross-checks, and authenticates information relating to ethnic and religious
affairs, and issues a decision to accept or reject the information;
- Operates without
re-examination or re-approval by the competent authority.
x
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
3
AI system for
automatically deciding to accept or reject applications for registration,
renewals, or cancellations of state management decisions in ethnic and
religious affairs
Assessments are
carried out only where the AI system:
- Automatically issues
the final decision to accept or reject applications for registration,
renewals, or cancellations of state management decisions in ethnic and
religious affairs;
- Operates without
examination, intervention, or re-approval by the competent authority.
x
4
AI system for
automatically issuing the final decision to approve or reject dossiers on determining
beneficiaries or areas eligible for ethnic minority policies
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
Automatically issues
decisions, including decisions that create, change, or terminate the lawful
rights and interests of individuals or communities.
x
5
AI system for
automatically executing the suspension of payment, recovery, or freezing of
the right to receive support due to suspected fraud
Assessments are
carried out only where the AI system:
- Automatically triggers
a measure interrupting the beneficiary's right to benefits or access to
assets or services;
- Operates without an
independent verification step before the measure takes effect.
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
6
AI system for
automatically giving order on the allocation or coordination of budget funds
or support policies according to an algorithm
Assessments are
carried out only where the system has the authority to execute or
automatically executes the budget allocation or support policy plan after the
decision on budget allocation or coordination takes effect
x
7
AI system for inferring
or classifying ethnicity, belief, or religion for the purpose of applying
management measures
Assessments are
carried out only where the AI system:
- Automatically creates classifications
of individuals or community groups based on ethnic, belief, or religious
attributes;
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
x
III
Health
1
AI-assisted surgery
system/surgical robot
Assessments are
carried out only to AI systems:
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
- Integrated into a
fully automated robot or intelligent robot performing tasks in place of human
actions.
x
2
AI-assisted surgery
system/surgical robot
(Machine and robotic
equipment systems using AI for automatic control)
Assessments are
carried out only where the AI system:
- Is integrated into a robot
directly executing treatment actions on the patient's body according to
medical orders;
- Operates without a
confirmation order from medical staff for each step of parameter change.
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
x
IV
Banking
1
AI system for
automatically executing electronic transactions in banking operations
Assessments are
carried out only to AI systems automatically executing electronic
transactions in banking operations where the following conditions are met:
- The system is capable
of automatically creating, modifying, confirming, and approving transactions;
- The transaction is of
high value as prescribed by the State Bank;
- There is no human
review, approval, or control before the transaction is executed.
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
2
AI-driven credit
extension system
Assessments are carried
out only to AI systems performing credit scoring, assessing repayment
capacity, and automatically issuing decisions on credit extension or
disbursement where the following conditions are met:
- The system issues the
final decision to approve or reject credit extension;
- The system's decision
is used directly for credit extension or disbursement without independent
approval by a credit officer;
- The loan is of a value
level prescribed by the State Bank.
x
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
Legal procedures
1
AI system for wide-area
biometric identification serving the resolution of public-interest civil cases
Assessments are
carried out only where the AI biometric identification system s widely
implemented at the location where the violation occurred or where the damage
occurred.
x
VI
Transport
1
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
Applies only to AI
systems:
- Controlling road,
rail, inland waterway, maritime, or aviation transport vehicles at a level that
does not require continuous human supervision or intervention during
operation;
- Or where the AI
system's decision serves as the direct basis for the vehicle to perform a
control action that could affect the safety of human life, health, or
property.
x
2
AI system controlling or
automatically operating signals and vehicle dispatch for road and rail
transport
Assessments are
carried out only to AI systems:
- Directly connected to
a road traffic control center or train dispatching system and having the
authority to automatically change signals, control cycles, organize traffic
flows, set running routes, or dispatch vehicles;
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
- Capable of directly
affecting traffic safety, train running safety, and road users.
x
3
AI system operating and
controlling the operation of key transport works and technical infrastructure
Assessments are
carried out only to AI systems:
- Having the right to
access and directly intervene the control and operational monitoring system
of road tunnels, major bridges, expressways, key transport hub works, or key
technical infrastructure systems including water supply, drainage, and urban
lighting.
- Where the AI system's
output is used for immediate execution or adjustment to the physical
operating state of the structure.
x
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
4
AI system automatically
detecting and assessing the risk of incidents and issuing orders to restrict
or suspend the operation of works or infrastructure structures
Assessments are carried
out only to AI systems:
- Automatically
assessing the safety condition of transport works and infrastructure,
construction works, or key technical infrastructure, and where the system's
results are used directly to decide the restriction of operation, suspension
of operation, isolation or adjustment to the operating state;
- Executed without a
compulsory independent human appraisal step.
x
5
AI system automatically
appraising, scoring, ranking, or classifying dossiers/application as the
basis for a final decision
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
- Used in state
management or the provision of public services within the scope of the
Ministry of Construction, where the output is used as the direct basis for
approving, rejecting, suspending, restricting, or changing the validity of
dossiers/applications, administrative procedures, licenses, certificates, or
operating conditions of organizations or individuals;
- Executed without an
independent review or approval step by a competent person.
x
6
AI system automatically
detecting violations or performing real-time biometric identification to
serve control and handling
Assessments are
carried out only to AI systems:
- Automatically
detecting violations, identifying subjects, analyzing images, sensor data, or
performing real-time biometric identification at traffic hubs or important
public works, where the results are used directly to handle violations,
enforce compliance, control security and order, restrict access, or apply
management measures;
- Executed without
independent verification by a competent officer.
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
7
Automated Air Traffic
Management (A-ATM) system issuing decisions on distribution of travel
direction and navigation of aircraft
Assessments are carried
out only where the AI system:
- Automatically issues
navigation commands, altitude changes, or distribution of travel direction
for aircrafts without direct supervision and approval by an air traffic
controller;
- Assessment does
not apply to systems that only provide suggestions or reference alerts.
x
8
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
Assessments are
carried out only where:
- The system directly
issues or executes decisions on aircraft separation, take-off/landing clearance,
or flight navigation;
- The system Operates
without independent approval by an air traffic controller.
x
9
AI system for Trajectory
Conflict Prediction & Resolution
Assessments are
carried out only where the AI system:
- Automatically detects
the risk of conflict and issues or executes an aircraft separation measure;
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
x
10
AI system for Air
Traffic Flow Management/Collaborative Decision Making (ATFM/CDM)
Assessments are carried
out only where the AI system:
- Is applied at an
international airport with a throughput exceeding 5 million passengers/year;
- Automatically
allocates and confirms slots without a review mechanism by an airport
coordinator. Assessments are carried out to DSS systems that only provide
suggestions.
x
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
AI-integrated flight
management system (Autopilot/FMS) capable of making independent navigation
decisions in abnormal situations
Assessments are carried
out only to AI systems integrated with autonomous equipment:
- The AI system is
capable of automatically adjusting the flight path, altitude, or handling
emergency situations without requiring the intervention and confirmation of
the pilot; does not include conventional autopilot systems that were type-certified
before the effective date of the Law on Civil Aviation.
x
12
New-generation AI-
integrated Collision Avoidance System (CAS) on aircraft
Assessments are carried
out only to AI-integrated CAS systems capable of automatically controlling
the aircraft to avoid collision without requiring pilot action. Assessments
do not apply to conventional TCAS/ACAS systems already type-certified before
the effective date of Law on Civil Aviation of Vietnam.
x
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
13
AI system predicting and
handling in-flight emergency situations (AI Emergency Flight Decision System)
Assessments are
carried out only where the AI system:
- Automatically proposes
or executes a decision to divert or make an emergency landing;
- Executed without
confirmation from the pilot or air traffic controller.
x
14
Aviation weather
forecasting AI system with the function of automatically issuing commands to
close a runway or cancel flights
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
- Automatically decides
to close a runway or cancel/delay a flight;
- Operates without
approval by the competent aviation meteorological authority.
x
15
Advanced Surface
Movement Guidance and Control System (ASMGCS Level 3-4)
Assessments are carried
out only where the AI system:
- Automatically grants
taxi clearance, allocates taxiways, or coordinates aircraft stands within the
airfield;
- Operates without
confirmation by a ground controller.
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
16
AI-integrated Stand/Gate
Allocation system
Assessments are carried out
only where the AI system:
- Automatically decides
the allocation of aircraft stands or aircraft boarding gates under actual
operating conditions;
- Operates without
approval by an air navigation officer.
x
17
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
Assessments are carried
out only where the AI system:
- Controls automatically
operating vehicles within the airside area, posing a risk of collision with
aircraft or airport infrastructure;
- Operates without
direct human operator supervision.
x
18
AI-integrated
U-space/Unmanned Aircraft System Traffic Management (U-Space/UTM) system
Assessments are carried
out only where the AI system:
- Automatically
navigates a UAV authorized to operate within an airport's air traffic control
zone or near a runway;
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
- Assessments do not
apply to Anti-UAV intrusion system.
x
19
Predictive Maintenance AI
system
Assessments are carried
out only where the AI system:
- Automatically issues a
grounding order or a decision permitting the aircraft to continue operating
based on AI analysis;
- Operates without
review and sign-off by a certified aviation engineer.
x
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
20
AI-integrated aviation
fuel management system (automatically ordering refueling)
Assessments are carried
out only where the AI system:
- Automatically
determines the quantity and timing of aircraft refueling;
- Operates without
confirmation by an aircraft operations engineer regarding fuel type.
x
21
Automated
(AI-integrated) Threat Detection system
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
- Automatically issues
warning decisions, classifies threats, triggers alarms, or isolates baggage;
- Operates without a
compulsory re-check step by aviation security staff; may pose risk in both
directions: missing dangerous items or triggering false alarms that disrupt
operations.
x
22
Behavior Detection AI
system in security control areas
Assessments are carried
out only where the AI system:
- Automatically
identifies suspicious behavior and triggers a security alert or intervention
measure;
- Operates without
initial confirmation by security staff; directly affects individual freedoms.
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
23
AI system for aviation
security management and automatically identifying risk subjects/refusing to
transport passengers.
Assessments are carried
out only where the AI system:
- Automatically refuses
to transport passengers or triggers a security measure;
- Operates without
verification by a competent aviation security officer.
x
24
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
Assessments are carried
out only where the AI system:
- Automatically executes
a countermeasure (signal jamming, shooting down, or taking control of an
unauthorized intruding UAV);
- Operates without a
confirmation order from aviation security forces or a competent military
authority.
x
25
AI system for automatically
controlling power flow and backup power at electrical stations/air traffic
control towers (SCADA/ICS)
Assessments are carried
out only to AI systems:
- Having the authority
to directly intervene in the physical control grid (SCADA/ICS) of important
national aviation infrastructure;
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
x
26
AI-integrated
Runway/Taxiway Incursion Detection System
Assessments are carried
out only where the AI system:
- Automatically detects
objects/vehicles/persons intruding into the airfield and triggers physical
infrastructure control measures (closing the runway, activating warning
lights);
- Operates without
confirmation by air navigation staff.
x
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
AI-integrated Foreign
Object Debris (FOD) Detection System
Assessments are carried
out only where the AI system:
- Automatically detects
foreign objects or safety hazards on the runway and triggers a warning or
operational restriction;
- Operates without
confirmation by airport operations staff.
x
28
AI system for monitoring
and controlling airport infrastructure (runway lighting, ILS/GBAS systems,
power supply)
Assessments are carried
out only where the AI system:
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
x
29
AI-integrated Instrument
Landing System/Ground-Based Augmentation System (ILS/GBAS with AI)
Assessments are carried
out only to landing systems for low-visibility conditions with AI integrated
for automated decisions; directly affecting aircraft landing safety.
x
30
AI-integrated security
risk assessment system for applications for aviation security clearance passes
...
...
...
Hãy đăng nhập hoặc đăng ký Thành viên
Pro tại đây để xem toàn bộ văn bản tiếng Anh.
x
31
AI system for
automatically inspecting, examining, and warning of violations of airport obstacle
limitation surfaces via satellite/aerial imagery data
Assessments are carried
out only where the AI system: is used as the legal basis for deciding to
suspend a work in violation of obstacle limitation surfaces or to impose
administrative penalties without on-site assessment.
x