This Tiny Island Is Giving Every Citizen Free ChatGPT. What does America do here?

Malta and Singapore have built the world's first real experiments in nudging entire populations into AI fluency. The hard question is whether the United States, which invented behavioral nudging, can do the same thing on its own soil.

In 2008, Cass Sunstein and Richard Thaler published Nudge, a book that quietly rewired how governments think about citizen behavior. The argument was modest in language and radical in implication. Most people will do the smart thing if you make the smart thing easy. Auto-enroll workers in retirement plans. Default organ donors to opt-out. Put the salad at eye level.

In the years since, Sunstein has spent much of his time on a more uncomfortable version of the same idea. In "Sludge and Ordeals" and in his earlier paper "Nudging and Choice Architecture: Ethical Considerations", he keeps returning to one constraint. Nudges work morally only when the choice architect's interests are aligned with the citizen's. When they aren't, you don't have a nudge. You have a funnel.

Eighteen years after the original book, governments are quietly applying the playbook to the most consequential skill of the decade. The question is no longer whether citizens should learn to use AI. It is how far a government can or should go to make sure they do, and whose interests get advanced when they try.

The Malta experiment

Malta has the cleanest example. In May 2026, the Maltese government signed deals with OpenAI and Microsoft to give every resident free access to ChatGPT Plus and Copilot for a year. The catch was small. Complete a short government course called "AI for All," delivered online through Malta's national digital identity system. OpenAI announced the partnership directly on its own site, and the Times of Malta covered the signing and the original budget proposal that preceded it.

Here is what the user experience is likely to look like once it rolls out. Imagine a schoolteacher in Sliema opening her phone, logging into the eID app she already uses for taxes and medical records, and seeing a single banner: Complete AI for All. Get ChatGPT Plus and Copilot, free for 12 months. The course is short, modular, and free. No separate account, no payment screen, no email verification loop. By the time she finishes, the subscription is provisioned to her existing identity.

Friction: near zero. Default state: not enrolled, but the path is short, social, and visible.

This is a textbook nudge. No one is forced. No one is fined. The state has simply made one option, learn AI and get the tool, dramatically more attractive than the alternative, which is doing nothing and paying twenty dollars a month yourself. It also, conveniently, hands OpenAI and Microsoft something they have never had before. A government-distributed customer acquisition channel for a consumer product.

The Singapore parallel

Three months before Malta's deal made headlines, Singapore quietly embedded the same logic into its national budget. The February 2026 Budget offered Singaporeans six months of free premium AI tool access, including ChatGPT and Gemini, after completing selected courses on the redesigned SkillsFuture platform. Prime Minister Lawrence Wong framed AI as a national strategic advantage, with a target of 100,000 "AI-savvy workers" by 2029.

The packaging is workforce development. The underlying mechanic is identical to Malta. Lower the cost of starting. Raise the cost of not starting. Let the population sort itself.

The two countries are notable not because they are large but because they are small. Malta has roughly 430,000 adult residents, fewer people than the city of Tampa. Singapore's workforce is around three million. Neither faces the legislative gridlock that slows similar proposals in the US, Germany, or Japan. In policy terms, they are speedboats, and right now, the speedboats are setting the pace.

The wider field

The pattern is broader than two countries. The UAE, in partnership with Google, launched "AI for All" in October 2025, targeting Emiratis across every age and profession with bundled access to premium Google tools, and extended free Gemini Pro subscriptions to every university student in the country. The UK has confined its push to civil servants through its AI Skills Boost program. Kazakhstan has committed to training one million people in AI skills by 2030.

What connects these programs is not geography or income level or political system. It is urgency.

The American question

This brings us to the harder problem. Could the United States, the country that invented behavioral nudging and exports more AI than the rest of the world combined, run a Malta-style program?

Probably not in the form Malta uses. The structural obstacles are real. There is no national digital identity in the US, so the friction-reduction trick that makes Malta's program work cannot be replicated at federal scale. Education and workforce policy are mostly state-run, which means any national course standard would either be voluntary and ignored, or mandatory and litigated. A federal subscription giveaway to OpenAI and Microsoft would be attacked from the left as corporate welfare and from the right as state-directed indoctrination, probably the same week.

But the US has done analogous things before. The GI Bill skilled an entire generation. Rural electrification wired the country in two decades. The Cooperative Extension System, run through more than 100 land-grant universities since the Smith-Lever Act of 1914, is the closest American analog to what Malta is doing: federally funded, locally delivered practical-skills training that most Americans have never heard of but that reaches into every county in the country.

So what would an American "AI for All" actually look like, if it were designed by someone who has read Sunstein and has actually worked inside a federal agency?

Roughly this:

1. Public libraries as the front door. Library cards are the closest thing the US has to a universal credential. Every major library system already runs digital literacy programs. Pay Anthropic, OpenAI, and Google a per-card fee for capped Plus-tier access through library accounts. Citizens enroll where they already are.
2. A federal tax credit for completed AI training, structured like the R&D credit. Employers get the credit when an employee finishes a certified course, not when they enroll. Completion is what changes behavior. Enrollment is what gets gamed.
3. Vendor neutrality by default. No single subscription deal. Citizens choose between two or three providers at the credential step. This makes procurement survivable and prevents the program from becoming OpenAI's loyalty card.
4. A Department of Labor apprenticeship category for AI-augmented roles, paired with funding through the WIOA system, which already serves more than three million Americans a year through state workforce boards. Apprenticeships are bipartisan and pre-existing. They route around the K-12 fights that would otherwise eat any federal curriculum proposal.
5. State pilots in California, Texas, and New York, designed to compete and publish results. The states are the laboratories. The Cooperative Extension model worked because Iowa did not have to wait for Washington.

Would it work? Partially. The American advantage is scale. Private actors will fill gaps the state will not, and they often do. The American disadvantage is that the gaps are exactly where the digital divide already lives. Malta will likely achieve something close to universal AI access in a year. The US, on this trajectory, will not get there in a decade.

The deeper question is whether replicating Malta would even be desirable.

The skeptic's case, taken seriously

The honest critique of these programs is not that they are coercive. They aren't. The critique is that the choice architect has interests that no longer line up cleanly with the citizen's.

When a government nudges citizens toward retirement savings, the state has no equity stake in the index fund. When it nudges them toward vaccination, no pharmaceutical company has a long-term subscription relationship with the citizen's calendar. When it nudges them toward ChatGPT Plus, OpenAI gains a state-acquired user with a behavioral pattern, a memory profile, and a likely renewal at full price the moment the free year ends. The citizen learns a useful skill. The vendor acquires a customer. The state, having brokered the introduction, is now structurally invested in the partnership succeeding.

This is not a reason to oppose AI literacy programs. It is a reason to design them with the question Sunstein keeps asking. Who benefits if the nudge works, and who is paying for it to work? Programs that answer those questions transparently, with disclosed commercial terms, vendor neutrality where possible, and sunset clauses where not, will age well. Programs that don't will look, in five years, less like nudges and more like the most successful customer acquisition campaign in the history of software.

The kicker

The countries running these experiments first will learn something the rest of us are about to need. How to teach an entire population a skill that did not exist five years ago, using infrastructure that did not exist ten years ago, in partnership with companies that may not exist in fifty.

Malta and Singapore are the lab. The United States, for now, is the audience.