Thailand is using AI to test whether its government can work as one

Thailand is trying to make five government agencies answer a citizen as if they were one.

The public AI prototype, Pathumma Connect, sends a question to five agency chatbots, combines their responses and returns one answer with citations. For that answer to be useful, each agency needs current public information that the system can search and combine with the others. The project therefore depends as much on the government machinery beneath the chatbot as on the interface itself.

What happened

Thailand’s Office of the Public Sector Development Commission (OPDC) has opened the service to connect the Department of Lands, Revenue Department, Department of Provincial Administration, Food and Drug Administration, and Rights and Liberties Protection Department.

The published flow starts at an application programming interface (API) gateway. An orchestration layer sends the question to relevant agency chatbots. A synthesis layer then combines their responses into one answer with sources.

Four connections were active when the portal was checked on 25 August 2026. The Revenue Department was marked as under maintenance or still being connected. The service remains a prototype for retrieving public information. The published material does not establish links to internal databases, live personal records or government transactions.

The institutional roles reflect the layers involved. OPDC provides policy direction and coordinates the agencies. The National Science and Technology Development Agency and its National Electronics and Computer Technology Center provide the Pathumma technology. The Digital Government Development Agency (DGA) connects the project to shared digital infrastructure and government AI principles.

At the 6 August 2026 launch event, Pakorn Nilprapunt, deputy prime minister for legal affairs, said government data should be accurate, current, standardised and interoperable. He also identified security, data protection, transparency, auditability, and technological and data sovereignty as requirements. These are stated requirements for the wider effort.

What it means

Consider a hypothetical property question. A citizen asking what they need to do to buy and register property could need information from the Department of Lands, Revenue Department and relevant local authority. The citizen sees one problem; government sees several agencies. Pathumma Connect is attempting that kind of coordination through one interface for its five pilot agencies.

Government information is organised around institutions. Agencies may use different terms for the same subject, keep forms and eligibility definitions in separate systems, publish guidance for different administrative contexts and update it on different schedules. A citizen can be told which office to ask next. The AI is being asked to do that administrative coordination first.

That can make existing information problems newly visible. A polished answer may still be poor if one source is outdated or two agencies describe the same procedure differently. Interoperability begins with agencies describing public information consistently enough for one system to search and synthesise it.

The portal warns that its answers are AI-generated, may be inaccurate, incomplete or outdated, and are non-binding. Users are told to verify the information with the relevant agency. Thumbs-up and thumbs-down feedback and OPDC contact details offer ways to flag a problem.

Legal authority still sits with the responsible agency. If Revenue guidance were to conflict with another agency’s guidance, for example, the system could cite both sources. Officials would need to determine which guidance governs, correct the public information and escalate unresolved cases through an established route.

The public record does not reveal a pilot-specific procedure for human review before an answer appears, correction and re-notification after an error, escalation to a named official, appeal, or governance of audit logs. Generic guidance from Thailand's Digital Government Development Agency (DGA) discusses human oversight, audit and traceability, but the record does not show how this prototype applies it.

Separately, a DGA and Microsoft training programme conducted in May 2025 attracted 2,017 applicants; 792 passed the first stage, 41 teams submitted AI proposals and 12 advanced to prototype development. That shows willingness inside government to experiment. Keeping a shared answer useful requires a different capability: agencies must update sources, reconcile conflicting guidance, monitor the service, correct errors and escalate disputes.

The consequential issue is whether Pathumma Connect remains a chatbot project or becomes an administrative one. If Thailand has to standardise its information to make the system work, the AI may end up changing government before it changes how citizens talk to government.

That makes Thailand a contained model other governments can study: start with public information, orchestrate existing agency chatbots through APIs, expose the sources and observe where institutional differences break the experience.

The lesson may not be the particular AI model Thailand uses. It may be the administrative work required before any model can reliably sit between a citizen and the state.

What to watch

• Does the prototype become an operational service? Watch for production status, stable availability across all five agencies and the addition of further public bodies.

• Does the system move from information to transactions? Authenticated records, personal data or government transactions would introduce a different level of legal, security and operational demands.

• Can government maintain one answer across five agencies? Look for a named owner when sources conflict, update timestamps, correction procedures, escalation routes and evidence that changes to an agency’s source information propagate into the shared answer.