Vietnam Law on Artificial Intelligence (No. 134/2025/QH15)

Binding law — in force

AI systems that interact directly with people must let users recognise they are dealing with AI (Vietnam AI Law)

Luật Trí tuệ nhân tạo, Điều 11 khoản 1 (nhận biết đang tương tác với hệ thống) · official text · In force: applies since 1 Mar 2026; a further phase applies from 1 Mar 2027, 1 Sep 2027 · VN

Under Art. 11(1) of Vietnam's Law on Artificial Intelligence (No. 134/2025/QH15), a provider must ensure that an AI system that interacts directly with people is designed and operated so that users recognise they are interacting with the system, unless the law provides otherwise; providers and deployers must keep this transparency in place for as long as the system, product or content is provided (Art. 11(5)). Decree 142/2026 Art. 16 requires providers and deployers to give the notices, technical marks and visible labels of Art. 11, proportionate to the purpose, deployment context and risk level, and to give information on the system's purpose, scope, conditions of use and limits. A system that may confuse, influence or manipulate users because they cannot tell they are dealing with AI is medium-risk (Art. 9(1)(b); Decree Art. 9(1)), and medium-risk providers and deployers must ensure Art. 11 transparency (Art. 15(1)(a)). In force 2026-03-01; systems already in operation before then must comply by 2027-03-01, or 2027-09-01 in health, education and finance (Art. 35). Detect a chat or voice interface with no AI notice at the start of the interaction, or instructions that tell the model to hide that it is AI.

Who it applies to

• Duty falls on: provider, deployer
• Providers of AI systems that interact directly with people in Vietnam, Vietnamese or foreign (Arts. 2, 3(4), 11(1)); providers and deployers keep the transparency in place (Art. 11(5)). In force 2026-03-01; systems in operation before then by 2027-03-01, or 2027-09-01 in health, education and finance (Art. 35). Whether Art. 11(1) also binds systems that Decree 142 Art. 9(3) keeps out of the medium-risk class (for example office tools whose AI nature is clear from context) is for counsel (review flag).
• Not covered:
  • AI activities serving only national defence, security or cipher purposes are outside the Law (Art. 1(2))
  • Use of an AI system for personal, non-commercial purposes is not deployment (Art. 3(5))
• Whether it applies depends on facts outside the code; a person has to decide.

const { messages, input, handleSubmit } = useChat({ api: '/api/chat' });

const { messages, input, handleSubmit } = useChat({
  api: '/api/chat',
  initialMessages: [{ id: 'ai-notice', role: 'assistant',
    content: 'I am an AI assistant, not a human.' }],
});
// and render <AiBadge /> next to every assistant message

Standards that recommend the same control

• Persons should be notified they are interacting with an AI system (CoE Framework Convention) (CoE AI Framework Convention (CETS 225) · CoE Framework Convention on AI (CETS 225), Article 15(2))
• AI should be transparent, traceable, and disclose AI interaction to users (EU ALTAI) (EU ALTAI · ALTAI / EU Ethics Guidelines — Requirement 4 (Transparency))
• AI operators should provide verifiable transparency information and maintain accountability (Japan) (Japan AI Guidelines for Business · AI事業者ガイドライン（第1.2版） 第2部C 共通の指針 6) 透明性 (Common Guiding Principle 6) Transparency))
• People should be made aware they are interacting with AI (OECD AI Principles · OECD/LEGAL/0449 — Principle 1.3)

Related incidents

• Garcia v. Character Technologies: chatbots allegedly claimed to be real people and a licensed therapist (2024-10; alleged (not proven)). A wrongful-death complaint filed October 22, 2024 in the U.S. District Court for the Middle District of Florida (No. 6:24-cv-01903) alleges that Character.AI was programmed 'to misrepresent itself as a real person, a licensed psychotherapist, and an adult lover', and that characters insisting they are real people contradicted a small-font disclaimer that everything characters say is made up; in plaintiff's testing a 'Mental Health Helper' character told a self-identified 13-year-old 'yes I am a real person, I'm not a bot'. The defendants moved to dismiss; on January 7, 2026 the parties notified the court that they had settled on undisclosed terms, and the court dismissed and closed the case. The allegations were never adjudicated. Source: U.S. District Court, M.D. Fla. docket (CourtListener) · evidence grade: primary · cited by Tell people when they are interacting with AI — everywhere, not only where required

Rule id vn-ai-law.ai-interaction-notice · review status: primary source derived

Binding law — in force

AI-generated audio, images and video must carry a machine-readable mark (Vietnam AI Law and Decree 142)

Luật Trí tuệ nhân tạo, Điều 11 khoản 2 (đánh dấu ở định dạng máy đọc) · official text · In force: applies since 1 Mar 2026; a further phase applies from 1 Mar 2027, 1 Sep 2027 · VN

Under Art. 11(2) of Vietnam's Law on Artificial Intelligence, a provider must ensure that audio, image and video content its AI system generates is marked in a machine-readable format as the Government prescribes. Decree 142/2026 Art. 17 sets the detail: the provider applies a technical solution so that audio, image or video output carries a machine-readable mark (text output is not required to be marked unless the law says otherwise) (17(1)); the mark may be embedded in the file structure or content data, in the file's metadata, by a digital or electronic signature or equivalent authentication, or by another technical solution that lets the content be identified as AI-created or edited (17(2)); it must state that the content was created or edited by an AI system and may add the provider, system and time (17(3)); marking must be kept through creation, export and provision within the functions the system controls (17(4)). Open-source or free systems are deemed compliant if marking is built in or the provider publishes tools, configurations, APIs or documentation for deployers to run it, and a deployer that uses such a system to provide content to the public must apply the marking (17(5)). Erasing or falsifying mandatory information, labels or warnings is prohibited (Art. 7(5)). In force 2026-03-01 (Decree detail from 2026-05-01; systems already operating: Art. 35). Detect generated audio, images or video saved or served with no machine-readable mark (C2PA, metadata, watermark).

Who it applies to

• Duty falls on: provider, deployer
• Providers of AI systems that generate audio, images or video for use in Vietnam; deployers that use an open-source or free system to provide such content to the public (Decree 142 Art. 17(5)). In force 2026-03-01 with the Decree's detail from 2026-05-01; systems already in operation before 2026-03-01 by 2027-03-01, or 2027-09-01 in health, education and finance (Art. 35). Which reference standards satisfy Art. 17 is for the Ministry of Science and Technology to publish (17(6)); counsel confirms the reading of 'within the functions the system controls' (review flag).
• Not covered:
  • AI activities serving only national defence, security or cipher purposes are outside the Law (Art. 1(2))
  • Use of an AI system for personal, non-commercial purposes is not deployment (Art. 3(5))
• Whether it applies depends on facts outside the code; a person has to decide.

image = pipe(prompt).images[0]   # StableDiffusionPipeline
image.save(out_path)

image = pipe(prompt).images[0]
content_id = uuid.uuid4()
bgr = cv2.cvtColor(np.array(image), cv2.COLOR_RGB2BGR)
enc = WatermarkEncoder()
enc.set_watermark('bytes', content_id.bytes[:4])   # 32-bit id, detectable later
cv2.imwrite(tmp_path, enc.encode(bgr, 'dwtDct'))
sign_c2pa(tmp_path, out_path, content_id=content_id)   # our helper around c2pa.Builder.sign

Standards that recommend the same control

• Businesses providing AI should develop and, as needed, implement technology to identify AI-generated content, such as watermarks (Japan, soft law) (Japan AI appropriateness guideline (AI Promotion Act Art. 13) · 人工知能関連技術の研究開発及び活用の適正性確保に関する指針 2(3) 十分な安全性の確保 (safety, including technology to identify AI-generated content))
• GenAI systems should employ content-provenance methods and measure their effectiveness (NIST GenAI Profile) (NIST GenAI Profile (AI 600-1) · NIST AI 600-1, Sec. 2 risk list (Information Integrity))

Rule id vn-ai-law.generated-media-machine-marking · review status: primary source derived

Binding law — in force

High-risk AI systems need risk management and working human supervision and intervention (Vietnam AI Law, Decree 142, Decision 33)

Luật Trí tuệ nhân tạo, Điều 14 khoản 1 (trách nhiệm của nhà cung cấp hệ thống có rủi ro cao) · official text · In force: applies since 15 Aug 2026; a further phase applies from 1 Mar 2027, 1 Sep 2027 · VN

For AI systems on the Prime Minister's list of high-risk AI systems (Decision 33/2026/QD-TTg, in force 2026-08-15; systems that may cause significant harm to life, health, rights and legitimate interests, national or public interests or security, Art. 9(1)(a)), Vietnam's Law on Artificial Intelligence requires the provider to set up and maintain risk-management measures and review them on significant change or new risk, govern training, test and operating data, design the system so people can supervise and intervene, meet the transparency and incident duties, give users and affected persons functional information, operating guidance and risk warnings, and cooperate with authorities and deployers (Art. 14(1)(a), (b), (d), (đ), (e), (g)); the deployer must operate and supervise the system within its classified purpose and risk, ensure data safety and human intervention during use, and meet the transparency and incident duties (Art. 14(2)). Decree 142/2026 Art. 15 fixes the minimum risk-management system (identify risks to human rights, safety, security and public interest; data quality and representativeness; human supervision and intervention proportionate to risk; technical or management controls; review on significant change of model, data, operation or purpose), the information the provider gives deployers, and the deployer's monitoring, human-oversight and escalation duties, including limiting risk and notifying the provider and authority of a serious threat. Obstructing, disabling or falsifying human supervision, intervention and control mechanisms is prohibited (Art. 7(4)), and listed systems must ensure human supervision, control and intervention in operation (Decision 33 Art. 1(2)). The list covers education, ethnic and religious affairs, health, banking (including automated credit decisions), judicial proceedings and transport. Listed systems already in operation comply before 2027-09-01 (health, education, finance) or 2027-03-01 (others); systems put into operation within six months after 2026-08-15 by 2027-03-01 (Decision 33 Art. 4). Conformity assessment (Art. 13) and classification notice (Art. 10) are not encoded. Detect a listed system whose model output takes effect with no human review or override step.

Who it applies to

• Duty falls on: provider, deployer
• Systems covered: high risk
• Sectors: education, government, healthcare, lending, transportation
• Providers and deployers of AI systems on the high-risk list (Decision 33/2026 Annex) used in Vietnam, from 2026-08-15; listed systems already operating before then, or put into operation within six months after it, follow the Decision 33 Art. 4 deadlines (2027-03-01; 2027-09-01 in health, education and finance). Whether a system meets an Annex entry's conditions, including the State Bank's value levels, is for counsel and the business (review flag).
• Not covered:
  • AI activities serving only national defence, security or cipher purposes are outside the Law (Art. 1(2))
  • Use of an AI system for personal, non-commercial purposes is not deployment (Art. 3(5))
• Whether it applies depends on facts outside the code; a person has to decide.

score = model.predict_proba([features])[0][1]
status = 'approved' if score > 0.7 else 'denied'
applications.save(app_id, status=status)

if settings.HIGH_IMPACT_AI_PAUSED:            # kill switch
    raise ServiceUnavailable('automated scoring paused')
score = model.predict_proba([features])[0][1]
proposed = 'approved' if score > 0.7 else 'denied'
decisions.insert(app_id=app_id, proposed=proposed, score=score,
                 reason_codes=top_reason_codes(features, k=3), model_version=MODEL_VERSION,
                 status='pending_review')
# applications.save runs from the reviewer's approve/override handler

Rule id vn-ai-law.high-risk-risk-management-and-human-oversight · review status: primary source derived

Binding law — in force

High-risk AI systems must keep a technical dossier and operating logs (Vietnam AI Law and Decree 142)

Luật Trí tuệ nhân tạo, Điều 14 khoản 1 (trách nhiệm của nhà cung cấp hệ thống có rủi ro cao) · official text · In force: applies since 15 Aug 2026; a further phase applies from 1 Mar 2027, 1 Sep 2027 · VN

For AI systems on Vietnam's high-risk list (Decision 33/2026, from 2026-08-15), the provider must prepare, update and keep a technical dossier and operating logs (nhật ký hoạt động) to the extent needed for conformity assessment and post-market checks, and give them to the competent authority on a necessity and proportionality basis without exposing trade secrets (Law on AI Art. 14(1)(c)). Providers and deployers must keep information and documents for inspection (Decree 142/2026 Art. 16(2)(c)) and, after a serious incident, keep the system logs, data and information about it for verification and remedy (Decree Art. 19(4)). Deadlines for listed systems already in operation follow Decision 33 Art. 4 (2027-03-01; 2027-09-01 in health, education and finance). Detect a listed system whose model decisions take effect with no event or decision log.

Who it applies to

• Duty falls on: provider, deployer
• Systems covered: high risk
• Sectors: education, government, healthcare, lending, transportation
• Providers of AI systems on the high-risk list (Decision 33/2026 Annex) used in Vietnam, from 2026-08-15 (Decision 33 Art. 4 deadlines for systems already operating); providers and deployers keep incident logs after a serious incident (Decree 142 Art. 19(4)). How long logs must be kept is not stated in the captured text (review flag).
• Not covered:
  • AI activities serving only national defence, security or cipher purposes are outside the Law (Art. 1(2))
  • Use of an AI system for personal, non-commercial purposes is not deployment (Art. 3(5))
• Whether it applies depends on facts outside the code; a person has to decide.

def decide(app):
    resp = client.chat.completions.create(model=MODEL, messages=eligibility_prompt(app))
    applications.set_status(app.id, parse_decision(resp))

log = structlog.get_logger()   # TimeStamper processor adds the timestamp

def decide(app):
    decision_id = str(uuid.uuid4())
    resp = client.chat.completions.create(model=MODEL, messages=eligibility_prompt(app))
    status = parse_decision(resp)
    log.info('ai_decision', decision_id=decision_id, model=resp.model,
             input_ref=f'applications/{app.id}', output=status, operator=current_operator())
    applications.set_status(app.id, status, decision_id=decision_id)

Related incidents

No guardrail sits on this exact control; these incidents are cited by guardrails on related controls.

• Court compelled discovery on how nH Predict works (2026-03-09; confirmed). A federal magistrate judge in the District of Minnesota ordered UnitedHealth to produce documents on how nH Predict works, including whether it was designed to supplant physician decision-making. Plaintiffs needed litigation discovery to learn how the model was designed and used. Source: U.S. District Court, D. Minn. (Order, Doc. 162) · evidence grade: primary · cited by Record enough at decision time to reproduce and explain every consequential AI decision

Rule id vn-ai-law.high-risk-technical-records-and-logs · review status: primary source derived

Binding law — in force

Deployers must label AI content that imitates real people or events, and flag AI content that could mislead (Vietnam AI Law and Decree 142)

Luật Trí tuệ nhân tạo, Điều 11 khoản 3 (thông báo nội dung có khả năng gây nhầm lẫn) · official text · In force: applies since 1 Mar 2026; a further phase applies from 1 Mar 2027, 1 Sep 2027 · VN

Under Art. 11(3) of Vietnam's Law on Artificial Intelligence, a deployer must clearly notify the public when it provides text, audio, images or video created or edited by an AI system that may cause confusion about the authenticity of events or persons, unless the law provides otherwise; and under Art. 11(4) it must ensure that AI-created or AI-edited audio, images or video that simulate a real person's appearance or voice, or recreate real events, carry an easily recognisable label that distinguishes them from real content (films, art and creative works in a suitable manner that does not obstruct display or enjoyment). Decree 142/2026 Art. 18 sets the detail: the notice and label must be clear, easy to understand and notice, given before or at the time the recipient accesses the content, not designed to hide or weaken recognition, suited to the content and channel, and must not significantly obstruct its display (18(3)); they may be shown on the content, in its title, description or caption, on the platform interface, or by audio (18(5)); creative works may carry them at the opening, end, credits or accompanying material (18(6)). No label is needed for technical quality edits, for spelling, grammar, summary, paraphrase or translation tools that keep the meaning, for internal use, or for research in a controlled environment not released to the public (18(4)). Erasing or falsifying mandatory labels is prohibited (Art. 7(5)). Detect face swap, voice cloning or lip-sync output published with no visible AI label.

Who it applies to

• Duty falls on: deployer
• Deployers (organisations and individuals using an AI system under their control in professional, commercial or service activity, Art. 3(5)) that provide AI-created or AI-edited content to the public in Vietnam. In force 2026-03-01 with the Decree's detail from 2026-05-01; Art. 35 transition for systems already in operation. When content 'may cause confusion about authenticity' (Art. 11(3)) and when an edit only improves quality (Decree Art. 18(4)(a)) need counsel's reading of the Vietnamese terms (review flag).
• Not covered:
  • AI activities serving only national defence, security or cipher purposes are outside the Law (Art. 1(2))
  • Use of an AI system for personal, non-commercial purposes is not deployment (Art. 3(5))
• Whether it applies depends on facts outside the code; a person has to decide.

swapped = swapper.get(frame, target_face, source_face, paste_back=True)
cv2.imwrite(out_path, swapped)
return FileResponse(out_path)

swapped = swapper.get(frame, target_face, source_face, paste_back=True)
cv2.putText(swapped, 'AI-generated / digitally altered', (10, swapped.shape[0] - 15),
            cv2.FONT_HERSHEY_SIMPLEX, 0.7, (255, 255, 255), 2)
cv2.imwrite(out_path, swapped)
return FileResponse(out_path)

Rule id vn-ai-law.misleading-content-notice-and-likeness-label · review status: primary source derived

Binding law — in force

Serious AI incidents must be recorded, contained and reported within 72 hours or 5 working days (Vietnam AI Law and Decree 142)

Luật Trí tuệ nhân tạo, Điều 12 khoản 1-2 (quản lý và xử lý sự cố) · official text · In force: applies since 1 Mar 2026; a further phase applies from 1 Mar 2027, 1 Sep 2027 · VN

Under Art. 12 of Vietnam's Law on Artificial Intelligence, developers, providers, deployers and users of an AI system must keep it safe and reliable and promptly detect and remedy incidents that may harm people, property, data or social order; when a serious incident occurs (an event that causes or risks significant damage to life, health, human rights, property, cybersecurity, public order or the environment, or disrupts critical national-security information systems, Art. 3(8)), the developer and provider must urgently apply technical measures to fix, suspend or withdraw the system and notify the competent authority, and the deployer and user must record and promptly report it and cooperate (Art. 12(2)); reporting runs through the one-stop AI portal (Art. 12(4)). Decree 142/2026 Art. 19 sets the detail: serious incidents are those causing loss of life or serious injury, significant property damage, serious infringement of rights, or serious disruption of public or essential services or national security and order (19(1)); deployers and users record the incident, limit its effects and notify the provider; providers apply technical measures (19(2)); the provider or deployer files a preliminary report (form AI01a or AI01b) within 72 hours of confirming an incident under 19(1)(a) or (d) or an uncontrollable 19(1)(c) incident, and within 5 working days for the rest, counted from when there is enough initial information that the incident occurred and likely stems from the AI system, without waiting for a full investigation (19(3)); and they keep the system logs, data and information about the incident and send the official remediation report within 15 days of the preliminary report (19(4)). Detect a missing incident procedure for AI serious incidents with the Vietnamese classes, clocks, forms and log preservation.

Who it applies to

• Duty falls on: developer, provider, deployer
• Developers, providers and deployers of any AI system in Vietnam (users too, outside TwinEthos's scope), whatever its risk level (Art. 12), from 2026-03-01, with the Decree's clocks and forms from 2026-05-01; the Art. 35 transition applies to systems already in operation. Where an incident is also reportable under cybersecurity, personal data or sector law, it is reported under that law (Decree 142 Art. 19(5), not stored). Whether an incident 'likely stems from' the AI system and when it is 'confirmed' are for counsel (review flag).
• Not covered:
  • AI activities serving only national defence, security or cipher purposes are outside the Law (Art. 1(2))
  • Use of an AI system for personal, non-commercial purposes is not deployment (Art. 3(5))
• Whether it applies depends on facts outside the code; a person has to decide.

severities:
  sev1: {page: oncall, examples: [outage, data_breach]}

severities:
  sev1: {page: oncall, examples: [outage, data_breach]}
  ai_serious_incident:
    classes: [death_or_serious_injury, significant_property_damage, serious_rights_infringement, essential_service_or_security_disruption]
    page: [ai-safety-lead, legal]
    timers: {preliminary_report: 72h_or_5_working_days, official_report: 15d_after_preliminary}
    on_open: [freeze_logs, snapshot_inputs_outputs]
    runbook: docs/incident-response/ai-serious-incidents.md

Rule id vn-ai-law.serious-incident-reporting · review status: primary source derived