Transparency and Patient Communication in Clinical AI in Viet Nam: What Healthcare Organizations Should Explain, Document and Disclose
IMPLEMENTATION INSIGHTS
Transparency and Patient Communication in Clinical AI in Viet Nam: What Healthcare Organizations Should Explain, Document and Disclose
IMPLEMENTATION INSIGHTS
Last reviewed: 1 October 2026
A patient does not need to understand a neural network to receive safe care. But patients should not be left with a misleading impression about the role technology has played when an AI system has meaningfully affected how their care was delivered. Transparency in clinical AI is therefore not about explaining every technical detail. It is about making sure that the right people receive the right information, at the right time and in a form they can actually use.
For healthcare organizations, this creates practical questions. What should clinicians know? What should patients be told? When should AI use be disclosed? What belongs in a patient-facing explanation, a consent discussion, an interface, a policy or the medical record? And how much information is enough before transparency becomes so technical that it stops being useful?
These questions have become more important in Viet Nam. The Law on Artificial Intelligence No. 134/2025/QH15 has been effective since 1 March 2026 and establishes specific transparency responsibilities. Decree No. 142/2026/NĐ-CP, effective since 1 May 2026, further requires providers and deployers to implement transparency in a manner appropriate to the system’s intended purpose, deployment context and level of risk, including providing information on intended purpose, scope of application, conditions of use and system limitations.[1,2]
Clinical communication, meanwhile, remains governed by healthcare law. The Law on Medical Examination and Treatment No. 15/2023/QH15 gives patients rights to information, explanation and consultation concerning their health and the methods and services involved in care, and to make choices after receiving relevant information about their condition, expected results and possible risks.[3] The implementation challenge is to bring these frameworks together without turning every use of AI into a new administrative exercise—or, at the other extreme, allowing AI to disappear completely behind the clinical workflow.
Transparency, explainability, disclosure and informed consent are not the same thing
These concepts are related, but they answer different questions. Transparency concerns whether relevant information about an AI system, its purpose, role, performance, limitations and risks is available to the people who need it. Explainability concerns whether the basis or logic behind an output can be understood meaningfully enough for the relevant purpose. Disclosure concerns informing someone that AI is being used or providing particular information about that use. Informed consent is a clinical and legal process in which a patient agrees to an examination, treatment, intervention, research activity or other relevant activity after receiving the information required for that decision.
They overlap, but they are not interchangeable. A healthcare organization can be transparent about an AI-supported process without necessarily creating a separate consent form. Conversely, having a patient sign a generic “AI consent” document does not by itself make an implementation transparent, understandable or safe.
International regulatory guidance takes a similarly audience-specific approach. The joint FDA, Health Canada and MHRA principles treat transparency as communication of appropriate information about a machine-learning-enabled medical device to relevant audiences, recognizing that the information needed by a patient, clinician, healthcare organization or regulator will differ.[6]
Viet Nam’s AI framework creates different layers of transparency
The first important point is that Vietnamese law does not impose one identical disclosure rule on every AI use.
Article 11 of the AI Law requires the provider to ensure that an AI system interacting directly with a person is designed and operated so that the user can recognize that they are interacting with AI, unless another law provides otherwise.[1] This is particularly relevant to patient-facing chatbots, conversational assistants, voice systems, virtual intake tools and similar applications in which the patient is directly interacting with the AI. A healthcare organization deploying a third-party system should therefore verify that this transparency requirement is properly implemented in the service it uses, even though the statutory design-and-operation obligation in Article 11(1) is placed on the provider.
Decree No. 142 adds a broader transparency layer. Article 16 requires providers and deployers to perform applicable notification, technical-marking and display-labeling duties and to provide information on the AI system’s intended purpose, scope of application, conditions of use and limitations. It also makes clear that transparency should be proportionate to purpose, deployment context and level of risk, and that transparency and explanation must comply with personal-data and data law.[2]
For high-risk AI systems, Article 14 of the AI Law adds further obligations. Providers and deployers have specific management and accountability duties, including providing users and affected persons with publicly available information at the level of a description of system functions, how the system operates and risk warnings necessary for safe use, within the responsibilities assigned to each actor.[1] At the same time, transparency and accountability do not generally require disclosure of source code, detailed algorithms, training data, model parameters or protected business and technology secrets unless another law requires it.[2]
The practical principle is therefore meaningful transparency, not maximum disclosure.
AI-generated content has its own disclosure rules
The AI Law also addresses AI-generated or AI-modified content. When a deployer provides text, audio, images or video to the public and the content could create confusion about the authenticity of an event, person or source, clear notification may be required. Certain AI-generated or modified audio, images and video simulating a real person or recreating a real event must also carry a recognizable display label.[1,2]
Decree No. 142 specifies that notices and display labels should be clear, understandable and recognizable and generally appear before or when the recipient encounters the content. Importantly, the Decree provides specific exemptions from the display-label requirement for certain technical editing; spelling or grammar correction; summarization, paraphrasing or translation that does not distort the basic meaning of the original text; content used internally and not provided publicly; and content generated in controlled research, development or testing environments that is not released publicly.[2] These are specific labeling exemptions, not blanket exemptions from other transparency, data-governance or accountability duties that may apply.
For healthcare organizations, this distinction matters. An internally AI-assisted clinical note is not automatically governed in the same way as synthetic public-facing media, but that does not mean the organization can ignore clinical governance, privacy, professional review or accountability.
Not every use of clinical AI requires a separate patient consent form
Healthcare organizations should be particularly careful with this issue. The current AI Law does not establish a blanket rule requiring every patient to sign a separate AI-specific consent form whenever AI is used somewhere in their care. AI may support scheduling, image processing, documentation, workflow prioritization, quality control or clinical decision support, and these uses differ substantially in how directly they affect a patient.
The absence of a universal AI-specific consent requirement does not make AI irrelevant to clinical communication. The Law on Medical Examination and Treatment protects patients’ rights to receive relevant information and explanation about their condition and the methods and services involved in care and to make choices after appropriate consultation about expected results and possible risks.[3]
As a practical governance principle, not a separate statutory legal test, healthcare organizations can consider how materially the AI affects the patient’s care, understanding or choice. If an AI system operates in the background of an administrative process and does not meaningfully alter the clinical method, patient interaction, decision or risk being discussed, an additional bedside conversation may provide little value. If AI directly engages the patient, substantially shapes a diagnostic or treatment process, introduces an important limitation or forms part of a method about which the patient needs to make an informed choice, the case for specific communication becomes much stronger.
In some situations, AI-related information may fit naturally within an existing clinical explanation or consent process rather than requiring a new “AI consent.” Research, clinical trials, new techniques and other specially regulated activities may have separate ethics and consent requirements and should be evaluated under the relevant framework.
Communication should reflect what the AI actually does
A useful organizational approach is to assess the AI’s role in the patient journey rather than asking only whether the word “AI” appears somewhere in the technology.
Consider an AI tool used solely to optimize appointment scheduling. Appropriate governance, privacy and security still matter, but routine clinical disclosure is unlikely to carry the same significance as it would for a system directly communicating with the patient.
Now consider radiology AI that highlights suspicious findings for a radiologist. The patient may never interact with the software, and the radiologist may remain responsible for interpreting the examination. The organization should ensure that clinical users understand the tool and can explain its role where relevant, while deciding how patient communication should be handled based on the use case, healthcare obligations, the system’s regulatory status and organizational policy.
The situation changes again when a patient communicates with an AI system that asks about symptoms, provides triage guidance or answers health questions. In that setting, Article 11’s direct-interaction rule becomes particularly important: the provider must ensure that users can recognize that they are interacting with AI.[1] The healthcare organization deploying the system should verify that this works in practice and determine what additional clinical information, limitations and escalation instructions patients need.
If the system is classified as high-risk, the additional requirements applicable to high-risk AI must also be addressed.[1] The communication strategy should therefore follow the role and consequences of the AI in care, rather than applying the same script to every technology.
Patients usually need to understand the role of AI, not its architecture
A common mistake is assuming that transparency requires a technical explanation of the model. Most patients do not need to know the number of model parameters, neural-network architecture, optimization method or full training pipeline, and even clinicians may not need that level of information for routine use.
Patients are more likely to need practical answers. What does the AI do? What information does it use? Does it make a decision or support a healthcare professional? What important limitations does it have? Can it be wrong? Does someone review the result? What happens when professional judgment and the AI disagree? Is a person available if the patient has questions or concerns?
For a clinician-support system, an explanation might be as simple as:
“We use an AI-assisted tool to help identify findings that may need closer review. It does not replace the clinician’s assessment. Your clinician reviews the result together with your other clinical information, and the tool has limitations and can make errors.”
That may be considerably more useful than a technically sophisticated explanation that a patient cannot interpret. The wording must, however, reflect what the specific system actually does. An organization should not say that “a doctor always makes the final decision” if the deployed system includes automation that operates differently.
Clinicians need more information than patients do
Transparency should be audience-specific. A patient may need a concise explanation of the role of AI and the continuing role of healthcare professionals. A clinician using the system needs considerably more information.
Clinical users should understand the intended use, target population, required inputs, clinically relevant performance characteristics, important limitations, conditions under which performance may deteriorate, relevant failure modes and how outputs are expected to influence-or not influence-care. They also need to understand what should happen when an AI output conflicts with clinical judgment.
Quality and patient-safety teams may need information on validation, monitoring, incidents, subgroup performance and model changes. IT and informatics teams may need details on interfaces, logging, version control, cybersecurity and availability. Procurement, legal and governance teams need still different information about evidence, regulatory status, responsibilities, contractual controls and lifecycle change.
One transparency statement cannot meet every audience’s needs. The joint FDA-Health Canada-MHRA principles organize this problem around six practical questions: who needs information, why they need it, what should be communicated, where it should appear, when it should be provided and how it should be presented.[6] That is a useful implementation model for healthcare organizations in Viet Nam as well.
Transparency should include limitations, not only benefits
AI communication can easily become promotional. A hospital announces “advanced AI technology.” A vendor advertises “95% accuracy.” A patient hears that AI “improves diagnosis.” Each statement may have some basis, but taken alone it can create an incomplete picture.
Responsible communication should address clinically relevant limitations as well as expected benefits. If image quality affects performance, users should know. If the system was validated mainly outside Viet Nam and locally relevant evidence remains limited, that may matter to clinical and governance teams. If a generative system can produce unsupported statements, that limitation should be clear. If monitoring is still developing, the organization should not communicate greater certainty than the evidence supports.
International transparency principles similarly identify intended use, performance, limitations, known risks, biases, confidence intervals, data-characterization gaps and circumstances in which real-world inputs differ from development and validation conditions as information that may be important to relevant audiences.[6]
Transparency should help people understand not only what the system can do, but where confidence in it should stop.
Explainability should be clinically meaningful
Explainability should not be confused with safety. An attractive heatmap does not prove that a prediction is correct. A feature-importance display does not automatically make a model clinically understandable, and an explanation generated by another AI system may itself be unreliable.
The better question is what kind of explanation the person making or affected by a decision actually needs. For a clinician, this might involve understanding influential inputs, uncertainty, limitations or whether an output is consistent with the clinical context. For a patient, it may be enough to understand the role of AI, its important limitations and how a healthcare professional is using its output.
The objective is not to make every algorithm mathematically transparent. It is to provide enough meaningful information for people to use technology safely, maintain appropriate human judgment and make informed healthcare decisions.
Direct patient-facing AI requires particular care
Patient-facing conversational AI deserves special attention because the interface itself is part of the clinical interaction. A chatbot used to book appointments is not equivalent to one that asks about symptoms. A system offering general health information is not equivalent to one recommending urgent care or reassuring a patient that symptoms appear low risk.
Where a patient interacts directly with AI, Article 11 places responsibility on the provider to ensure that the system is designed and operated so the user can recognize that the interaction is with AI.[1] A healthcare organization deploying that system should verify the disclosure in practice and should also determine what information is needed about the scope of the service, important limitations, escalation and access to human assistance.
Organizations should ask practical questions. Can the patient reach a person when needed? What happens if the system encounters a high-risk symptom? Does it communicate uncertainty appropriately? Could general information reasonably be mistaken for individualized medical advice? A small “Powered by AI” notice may identify the technology without adequately explaining its clinical role.
The more useful question is whether patients understand who or what they are interacting with, what the system can and cannot do and when human assistance becomes necessary.
Generative AI creates a different transparency challenge
Generative AI may assist with clinical notes, letters, patient instructions, summaries, translations and educational material. The governance question is not simply whether every sentence touched by AI must carry an “AI-generated” label.
Accountability and review matter more. A clinician-reviewed draft is different from automatically generated material sent directly to a patient without meaningful professional review. Healthcare organizations should therefore define which AI-assisted outputs require human review, who approves patient-facing content, when statutory notification or labeling rules apply and which outputs should not be sent automatically.
For public AI-generated content, the specific notification and labeling requirements of the AI Law and Decree No. 142 should be assessed. As noted above, Decree No. 142 creates exemptions from the display-label requirement for defined technical editing, summarization, paraphrasing and translation that do not distort the original meaning, as well as certain internal and controlled R&D uses.[2] It does not turn those activities into blanket exemptions from broader clinical, privacy or accountability obligations.
The practical objective is not to label every internally assisted sentence. It is to avoid misleading patients about the source, reliability and level of human review of information that matters to their care.
Data transparency is a separate governance obligation
A healthcare organization can explain the clinical role of AI well and still have inadequate data governance. Transparency about the clinical role of AI and transparency about personal-data processing are related, but they are not the same obligation.
Viet Nam’s Law on Personal Data Protection No. 91/2025/QH15 and Decree No. 356/2025/NĐ-CP have been effective since 1 January 2026.[4,5] Where an AI system processes patient information, organizations should separately assess the purpose of processing, organizational and vendor roles, access controls, retention, security, cloud services, transfers and other applicable data-protection requirements.
This distinction becomes especially important where an external vendor or cloud service is involved. A patient may understand that AI supports their care and still have a separate interest in understanding how their personal data are being processed.
Decree No. 356 also establishes additional requirements for qualifying big-data processing involving personal data, including an appropriate mechanism to notify and explain to data subjects how their personal data are used in the big-data analytics system.[5] This should not be generalized to every AI implementation; it applies where the activity falls within the Decree’s definition and provisions on big-data processing.
Documentation matters, but documentation should be purposeful
Transparency should not depend entirely on whether an individual clinician happens to remember the right explanation. Healthcare organizations should maintain a consistent internal record of how transparency and patient communication will be handled for significant clinical AI systems.
That record can identify the intended use, relevant audiences, information supplied by the provider, known limitations, applicable patient-facing notices, responsibilities for answering questions, approved communication materials and circumstances requiring additional explanation. It can also define when the role of AI should be documented within the medical record.
A blanket organizational requirement to document every background use of AI in every medical record may add documentation without necessarily improving care. More targeted documentation may be particularly useful where AI materially contributed to a clinical decision, a specific disclosure or consent process was required, AI-generated information was communicated directly to the patient, a clinically significant disagreement between a healthcare professional and the AI affected care, or an AI-related incident or complaint occurred.
These are practical governance examples rather than a universal statutory documentation rule. Organizations should also assess any specific medical-record, medical-device, research or specialty requirements that apply to the particular use case.
Transparency information needs change control
Clinical AI changes. A system may receive a new model or software version, expand its intended use, connect to a new data source or move into a new clinical workflow. A healthcare organization may also change how strongly clinicians rely on its output.
Patient-facing and clinician-facing information can therefore become outdated even when it was accurate at launch. Article 11 of the AI Law requires providers and deployers to maintain applicable transparency information throughout the period in which the relevant system, product or content is provided to users.[1] Decree No. 142 also requires transparency to remain appropriate to intended purpose, deployment context and risk.[2]
Transparency materials should consequently form part of AI change management. When a material change occurs, the organization should ask whether user instructions, patient information, FAQs, website content, warnings, training or consent materials also need to change. Not every software patch requires a new patient notice, but communication should change when the information clinicians or patients need to understand or safely use the system has materially changed.
Patient questions and complaints need somewhere to go
Transparency is incomplete if an organization supplies information but provides no pathway for questions or concerns. Patients may want to know whether AI was involved, whether a healthcare professional reviewed the output, how their data were used or whether an AI-supported recommendation can be reconsidered. They may also identify an error that the organization had not previously detected.
Healthcare organizations should connect AI communication with existing patient-relations, clinical-governance, privacy and incident-reporting processes. Staff responding to questions should know where authoritative information is maintained and who can provide a more detailed clinical, technical or privacy explanation when needed.
An AI-related complaint should not automatically be treated as an IT problem. Depending on the issue, it may involve clinical judgment, patient safety, data protection, vendor performance, human factors or several of these at once.
WHO/Europe’s 2026 Knowledge Community identified strong governance, transparency and codesign with patients and clinicians among important enablers of responsible AI and identified participation by patients, communities and frontline professionals as one of the priority areas discussed by participants. These findings reflect the Knowledge Community’s views and are not a binding WHO regulatory standard.[7]
A practical Transparency and Patient Communication Gate
Before routine deployment, a healthcare organization can apply a simple Transparency and Patient Communication Gate:
Audience clarity: Have we identified who needs information about this AI—patients, caregivers, clinicians, administrators, regulators or other affected persons?
AI role clarity: Can we explain in plain language what the system does, where it sits in the care pathway and how its output can affect decisions or actions?
Human oversight clarity: Do users and, where relevant, patients understand what role healthcare professionals retain and what happens when human judgment and AI disagree?
Limitations and risk clarity: Are clinically important limitations, uncertainty, failure modes and conditions of use communicated to the people who need them?
Legal transparency readiness: Have applicable duties relating to direct AI interaction, general transparency, high-risk AI and AI-generated content been identified and implemented?
Patient communication readiness: Has the organization defined when AI should be explained during routine care, when relevant information belongs within an existing consent discussion and when additional AI-specific communication is unlikely to add value?
Data transparency readiness: Are clinical AI communication and personal-data transparency managed as connected but distinct obligations?
Documentation and change control: Are approved notices, explanations, responsibilities and communication materials documented, version-controlled and reviewed when the technology or use case changes?
This Gate is a practical organizational governance framework. It is not a statutory checklist under Vietnamese law and should be adapted to the technology, clinical use, risk classification and organizational context.
What this means for healthcare organizations in Viet Nam
Healthcare organizations do not need to explain every technical detail of every AI system to every patient. They do need to understand which systems affect patients, how they affect them and what information becomes important because of that role.
For patient-facing AI, the provider’s direct-interaction transparency obligation should be verified in the deployed service, and the healthcare organization should determine what additional clinical information and escalation pathways patients need. For high-risk AI, the additional transparency, risk-management and accountability requirements should be addressed explicitly. For clinician-support systems, clinical users need enough information to use the technology safely and explain its role where relevant. Where AI becomes relevant to the examination or treatment method, expected risks, patient choice or another information duty under healthcare law, organizations should determine how that information belongs within normal clinical communication.
The answer will rarely be a single hospital-wide sentence saying, “We use artificial intelligence.” That tells patients very little. A more useful approach is specific to the use case: explain what the AI does, what the healthcare professional still does, which limitations matter, what happens if concerns arise and where the patient can ask questions.
Transparency should also remain proportionate. Too little information creates opacity. Too much technical information can create confusion without improving understanding. For healthcare leaders, the important question is therefore not simply “Have we disclosed that we use AI?” It is “Have we given patients, clinicians and other affected people the information they actually need to understand the role, limits and accountability of this system?”
That is meaningful transparency. Responsible clinical AI should not require patients to become AI experts, but neither should it require them to trust technology simply because they were never given enough information to ask the right questions.
Professional education notice
This article is intended for professional education and knowledge exchange. It does not constitute individualized legal, regulatory, clinical, medical-device, personal-data or informed-consent advice.
The communication practices and Transparency and Patient Communication Gate described in this article are practical organizational recommendations. They are not a statutory checklist and should not be interpreted as stating that every use of AI requires individual patient notification, a separate consent form or documentation in the medical record.
Whether particular information must be disclosed, documented or incorporated into a consent process depends on the AI system, the organization’s legal role, risk classification, intended use, degree of automation, interaction with patients, clinical context, data processing and other applicable legal and professional requirements.
Healthcare organizations should separately assess requirements arising from the Law on Artificial Intelligence, the Law on Medical Examination and Treatment, personal-data protection legislation, medical-device regulation, research and ethics requirements and other applicable specialized legislation.
References
National Assembly of Viet Nam. Law No. 134/2025/QH15 on Artificial Intelligence. Adopted 10 December 2025; effective 1 March 2026.
Government of Viet Nam. Decree No. 142/2026/NĐ-CP Detailing a Number of Articles and Measures for Implementation of the Law on Artificial Intelligence. 30 April 2026; effective 1 May 2026.
National Assembly of Viet Nam. Law No. 15/2023/QH15 on Medical Examination and Treatment. Adopted 9 January 2023; effective 1 January 2024.
National Assembly of Viet Nam. Law No. 91/2025/QH15 on Personal Data Protection. Adopted 26 June 2025; effective 1 January 2026.
Government of Viet Nam. Decree No. 356/2025/NĐ-CP Detailing a Number of Articles and Measures for Implementation of the Law on Personal Data Protection. 31 December 2025; effective 1 January 2026.
U.S. Food and Drug Administration, Health Canada and UK Medicines and Healthcare products Regulatory Agency. Transparency for Machine Learning-Enabled Medical Devices: Guiding Principles. June 2024.
World Health Organization Regional Office for Europe. Report of the Knowledge Community on Responsible Artificial Intelligence in Health. 1 September 2026. WHO/EURO:2026-13381-53155-83077.