Thailand is about to spend 1.62 billion baht, roughly US$48 million, to hand as many as five million of its citizens a full year of paid-tier access to more than thirty generative models from over fourteen providers, including ChatGPT, Claude, Gemini and Grok. That is the headline number. But the more consequential design decision sits one layer beneath it: before a Thai citizen's prompt reaches a server in California, it is supposed to pass through a government-procured filter meant to strip out personal and sensitive information.
The programme, TH-AI Passport, is run by the Ministry of Digital Economy and Society through the Office of the National Digital Economy and Society Commission. Advance registration opened on 19 August and drew 307,425 sign-ups in half a day, enough to jam ThaID, the national identity-verification app. By 5pm on 20 August the count stood at 1,033,635. A fifth of the ceiling went in two days.
The dates have circulated inconsistently: 28 August is a formal ceremony, open to opposition politicians as well as industry. Public access to AiPASS, the aggregation platform at the centre of the scheme, begins on 31 August and runs to 31 August 2027. Eligibility is narrow: Thai nationals aged 15 or over, registering personally, no proxies, verified through ThaID or Thang Rath. The subsidy is also conditional. Users get about 30 days of open access, after which further credits must be earned by completing courses from a catalogue of more than 96 training modules, a design the ministry calls Learn to Earn. And the money follows usage, not registration: after a June climbdown, payments were restructured to a pay-per-active-user basis at 24.70 baht per active user per month, about 75 US cents.
The target is a statistic that stings in Bangkok. Thailand's AI usage rate sits at 10.7 percent, 89th globally and behind most of its Southeast Asian neighbours. The ministry wants it above 20 percent by 2027.
The screening layer, and what is documented
Officials have leaned hard on the data story. Personal and usage data, the government says, stays and is processed domestically, while content forwarded to foreign providers is screened for personal or sensitive information, and providers are barred from retaining it or training on it.
Digital Economy and Society Minister Chaichanok Chidchob has made that his core defence, arguing AiPASS is safer than an ordinary consumer subscription, where prompts leave the country unless the user disables the right settings. "I can confirm that this project is secure," he said as registration opened. "As stipulated in the contract, participating foreign AI companies cannot further process the data."
The platform's own privacy notice is more equivocal. It discloses collection of name, national ID number, occupation, income bracket, location, device data and the content of AI conversations including uploaded files, and states that conversation content is transmitted to overseas providers including OpenAI, Google, Microsoft and Anthropic. Thailand's National Electronics and Computer Technology Centre says it has asked providers not to use the data for training where possible, and that anonymisation measures are applied. Users are also advised not to type sensitive personal information into the chat at all, which is a curious instruction if the screening layer works as advertised.
The politics
None of this is landing in a calm room. The opposition People's Party has pushed to kill the project and taken a complaint to the National Anti-Corruption Commission. Rukchanok Srinork, who chairs the House committee on budget monitoring, has been blunt. "No one wants to obstruct Thailand's AI development," she said. "But there are irregularities throughout this 1.6-billion-baht project, and it is unclear why the government insists on pushing ahead."
Under scrutiny: a tender announced and bid inside 34 days, terms of reference critics say favoured large incumbents, a comparison with a rival state AI platform built for 2.4 million baht, and a registration site on a .net domain rather than the .go.th suffix used by government services. Chaichanok says all of it complied with the contract and the law. "Our intention is for scrutiny," he said. "This may not be pleasing to everyone, but it is legal."
Prime Minister Anutin Charnvirakul, backing the scheme from Canberra, frames it as labour protection. "We want Thai people to gain the widest possible access to AI, with fewer barriers, so they can strengthen their competitiveness and develop their capabilities through the most advanced technology available," he said.
Why It Matters
Strip away the procurement fight and what remains is industrial policy: the state as bulk purchaser, correcting an access gap the market produced. Premium tiers cost a Thai student a meaningful slice of monthly income; the government has negotiated them down to the price of a bottle of water per month. That is a real subsidy, and it flows overwhelmingly to American labs. Thailand is not building a national model here. It is buying seats.
The more exportable idea is the screening proxy. Sovereign AI has mostly meant sovereign compute, ruinously expensive and years away for middle-income countries. Thailand is testing a cheaper substitute: keep identity and usage data at home, insert a state-run intermediary between citizens and foreign inference, and push the sovereignty question into contract law and a filtering layer rather than into data centres. If it works, it is a template any country with a procurement office could copy. If the filter is unaudited and the training ban unverifiable, it is a sovereignty claim resting on a promise.
What to Watch
The number that matters after 31 August is not registrations but activations, and then survival past the 30-day credit cliff, since payment is now tied to active use. Watch whether the ministry publishes a specification or independent audit of the screening layer, which would turn the proxy model into something other governments could evaluate. Watch the NACC complaint and the Budget Committee hearings, which could stall disbursement regardless of uptake. And watch the 2027 target: doubling national AI usage in eighteen months is the metric the government picked, and the one it will be judged against.
“I can confirm that this project is secure. As stipulated in the contract, participating foreign AI companies cannot further process the data.”— Chaichanok Chidchob, Minister of Digital Economy and Society, Thailand