‘Other governments will be watching this closely’ — South Korea starts testing free AI for the public

Most governments still talk about artificial intelligence as something to regulate or subsidize.

<![CDATA[ <article> <p>Most governments still talk about <a href="https://www.techradar.com/news/what-is-ai-everything-you-need-to-know">artificial intelligence</a> as something to regulate or subsidize. South Korea is trying something considerably bolder. It wants to make advanced AI available free to every citizen, treating access less like a premium software subscription and more like a basic piece of national infrastructure.</p><p>The country's Ministry of Science and ICT has selected three consortiums, led by SK Telecom, KT and Kakao, for its <a href="https://www.koreatimes.co.kr/business/tech-science/20260828/skt-kt-kakao-consortiums-selected-for-free-ai-service-for-public" target="_blank">"AI for All" </a><a href="https://www.koreatimes.co.kr/business/tech-science/20260828/skt-kt-kakao-consortiums-selected-for-free-ai-service-for-public" target="_blank">program</a>. The plan is to give Koreans free, unlimited access to general-purpose AI services built primarily on domestic models, with beta testing before a full launch expected by the end of 2026. The government is supplying a combined 512 Nvidia B200 GPUs this year and plans to subsidize nationwide operating costs from 2027.</p><p>That already sounds like a government-backed ChatGPT competitor, but South Korea's ambitions go considerably further. The services are supposed to develop into agents capable of helping citizens find government programs, submit applications, make reservations and payments, and carry out other tasks. The government's longer-term ambition is effectively one AI agent for every citizen.</p><p>Plenty here could attract international attention, but the project’s real value may have little to do with giving everyone a free chatbot.</p><h2 id="far-more-than-a-chatbot">Far more than a chatbot</h2><p>South Korea has a strong argument for intervening. Its science ministry says roughly two-thirds of Koreans have used AI services, and around 23 million people have used generative AI. ChatGPT alone had an estimated 23.45 million monthly active users, with Gemini and Claude having far lower but still notable monthly users of 8.45 million and 2.41 million, respectively.</p><p>Those figures are impressive until you consider the people missing from them. If AI really does become as important as its advocates believe, the ability to pay $20 or $30 every month for the best models starts to matter. Free tiers help, but their limits, features, and even continued existence are decisions made by private companies.</p><p>South Korea's government has explicitly identified that dependence as a problem. It argues that citizens relying on free versions of foreign AI services remain vulnerable to usage restrictions and price hikes. Its answer is a publicly supported alternative built largely around Korean models, with domestic models required to account for at least half of the service. Plus, South Korea gets to widen AI access while supporting domestic model developers and reducing its reliance on foreign platforms. </p><p>“South Korea’s ‘AI for Everyone’ project rests on a principle worth defending: access to AI should not depend on what you can afford, and a country is stronger for having its own capability. The ambition is not in question. How it works in practice is,” said Nik Kairinos, CEO & Co-founder of <a href="https://www.raidsai.ai/" target="_blank">RAIDS AI</a>, which provides testing and detection of threats to AI services. </p><p>“When a publicly provided model or agent goes rogue, gives someone wrong information, or causes harm, who answers for it? The government that commissioned the service, the company that built it, or both? Citizens will not have chosen their provider, and many will not know which one they are using.”</p><figure class="van-image-figure inline-layout" data-bordeaux-image-check ><div class='image-full-width-wrapper'><div class='image-widthsetter' style="max-width:2560px;"><p class="vanilla-image-block" style="padding-top:56.25%;"><img id="mLi8aMeEEGaxUAdNrNXpVQ" name="shutterstock_2433515433 copy" alt="Seoul City at night South Korea" src="https://cdn.mos.cms.futurecdn.net/mLi8aMeEEGaxUAdNrNXpVQ.jpg" mos="" align="middle" fullscreen="" width="2560" height="1440" attribution="" endorsement="" class="inline"></p></div></div><figcaption itemprop="caption description" class=" inline-layout"><span class="caption-text">Seoul City at night, South Korea. </span><span class="credit" itemprop="copyrightHolder">(Image credit: Shutterstock / Kampon)</span></figcaption></figure><h2 id="combining-ai-with-public-services">Combining AI with public services</h2><p>I am skeptical about the idea of governments offering general-purpose AI subscriptions for basic technical reasons, as well. AI models improve at an absurd pace, and taxpayers would effectively be underwriting an endless technological race against OpenAI, Google, Anthropic, and whatever comes next.</p><p>“I think it is not rational to spend money on an unlimited general chatbot striving to compete with ChatGPT, Claude, or Gemini, since the commercial world of artificial intelligence changes very quickly,” said John Park, a founding membеr of technical staff at agentic AI startup <a href="https://theagi.company/" target="_blank">AGI, Inc</a>. </p><p>“Instead of that, a more appropriate strategy will be to combine AI with public services that people are already using, for instance, with the government’s call center, benefit applications, tax assistance, appointment booking, and document creation. That way, AI becomes part of the cost of providing public services, just like a website or a call center.”</p><p>That, to me, is the real blueprint hidden inside South Korea's experiment. Governments don't need to build the world's best chatbot. They need to make dealing with government better. An AI that knows which parental benefit you qualify for, finds the correct form, and explains any baffling sections before helping you submit it is useful in a way that justifies public spending. A do-anything chatbot lacks that justification</p><p>That's the strategy in South Korea. Kakao plans to put its service inside KakaoTalk. While KT has presented an agent spanning public services, education, finance and shopping, and SK Telecom plans phone- and text-based agents for those reluctant to engage with apps or websites.</p><p>You can see hints of similar ideas in other countries. The UK launched GOV.UK Chat in its government app in May, allowing people to ask natural-language questions about government information. Singapore's VICA conversational AI platform is used across more than 60 government agencies, and France announced plans for a public health chatbot and a conversational assistant. AI as a public utility probably makes sense when the "utility" is access to government itself.</p><figure class="van-image-figure inline-layout" data-bordeaux-image-check ><div class='image-full-width-wrapper'><div class='image-widthsetter' style="max-width:2560px;"><p class="vanilla-image-block" style="padding-top:56.25%;"><img id="e2szkPKKqJShKrZsRW2VZk" name="shutterstock_2379442805 (1) copy" alt="Crowds at the Gangnam district nightlife in Seoul." src="https://cdn.mos.cms.futurecdn.net/e2szkPKKqJShKrZsRW2VZk.jpg" mos="" align="middle" fullscreen="" width="2560" height="1440" attribution="" endorsement="" class="inline"></p></div></div><figcaption itemprop="caption description" class=" inline-layout"><span class="caption-text">Crowds at the Gangnam district. Nightlife in Seoul </span><span class="credit" itemprop="copyrightHolder">(Image credit: Shutterstock / f11photo)</span></figcaption></figure><h2 id="privacy-policy-for-freedom">Privacy policy for freedom</h2><p>The difficulty begins when an assistant stops pointing citizens toward services and starts acting on their behalf. A chatbot incorrectly describing a benefit is bad. An agent incorrectly applying for one, canceling something, making a payment, or supplying inaccurate information under your identity is an entirely different class of failure.</p><p>“Once an agent is transacting for a citizen inside public services, the bar for knowing what it did, and why, is far higher than for a system that only answers questions,” Kairinos pointed out. “Other governments will be watching this closely, because most of them will face the same choice within a few years. Widening access to AI is worth doing. It just cannot come at the expense of accountability, evidence and public trust.”</p><p>That warning becomes particularly important when considering what people actually tell AI assistants. We already happily type things into commercial chatbots that we might hesitate to tell acquaintances. Put an AI assistant between a citizen and the welfare system, tax authority, or health service, and the information becomes far more sensitive.</p><p>“Citizens utilizing AI for benefit applications and other services could disclose very personal details about themselves: health, income, family, and finances. Such information needs to be kept separate from other aspects like advertising, commercial profiling, and training of models,” Park said. “Citizens need to have access to such essential services without having to consent to the use of their data elsewhere. Otherwise, the free service may cost the citizenry nothing in monetary terms, but at the cost of their personal information.”</p><p>There is an uncomfortable detail in South Korea's own description of "AI for All" here. The science ministry has said participating companies will be expected to develop revenue models, giving the example of using prompt data collected while providing the services. That may help make free AI financially sustainable, but it also makes the boundaries around data use enormously important.</p><p>Citizens should not have to wonder whether describing their unemployment, disability, debt or family circumstances to obtain public assistance is also enriching a commercial profile or future training set. If AI becomes part of the machinery of government, privacy protections need to resemble those surrounding public records more than the terms and conditions of a consumer app.</p><figure class="van-image-figure inline-layout" data-bordeaux-image-check ><div class='image-full-width-wrapper'><div class='image-widthsetter' style="max-width:2560px;"><p class="vanilla-image-block" style="padding-top:56.25%;"><img id="hryVK6UojX4kCpTKtbSarg" name="shutterstock_1944700558 copy" alt="Panoramic view of Gogunsan Islands from Daejangbong peak in Gunsan, Korea" src="https://cdn.mos.cms.futurecdn.net/hryVK6UojX4kCpTKtbSarg.jpg" mos="" align="middle" fullscreen="" width="2560" height="1440" attribution="" endorsement="" class="inline"></p></div></div><figcaption itemprop="caption description" class=" inline-layout"><span class="caption-text">Panoramic view of Gogunsan Islands from Daejangbong peak in Gunsan, Korea </span><span class="credit" itemprop="copyrightHolder">(Image credit: Shutterstock / Sanga Park)</span></figcaption></figure><h2 id="public-utility-not-public-chatgpt">Public utility, not public ChatGPT</h2><p>South Korea may simultaneously prove that universal AI access is a good idea and that "free ChatGPT for everyone" is the wrong interpretation of it.</p><p>“South Korea is a successful example of how a nation can increase AI adoption, help local firms, provide better language-specific solutions, and decrease reliance on international websites. The U.S. would most likely develop a more fragmented system of AI services with different programs dedicated to tax, benefits, healthcare, education, and library applications,” Park said. </p><p>“Basic capabilities of AI, like search, translation, summarization, form filling, and help with public services, could turn out to be free or almost free. The main task of the government will be to ensure accessibility, privacy protection, transparency, and the right to human review, while private companies continue competing on more advanced services.”</p><p>That fragmented version is ultimately more convincing than the grander vision. There is an extremely strong argument for ensuring nobody is denied access to healthcare, benefits, or tax assistance because they cannot afford the AI tools increasingly used to navigate them.</p><p>Governments can guarantee a baseline of useful capabilities tied to citizenship and public services. Private companies can continue fighting ferociously over the frontier, selling better models, enormous context windows, creative tools, and more.</p><p>South Korea is attempting something much larger. It wants broad AI access, domestic technological sovereignty and eventually personal agents that can conduct parts of citizens' economic and social lives. That makes the country a very expensive live experiment rather than a blueprint. AI is becoming part of how the country thinks about economic infrastructure and public administration.</p><p>Other governments should copy the principle before they copy the product. Access to basic AI capabilities is likely to become important enough that leaving it entirely to monthly subscriptions will create another digital divide. Governments can ensure everyone benefits from the real value of AI tools, which will probably be the most boring examples and come with strict data rules.</p><p>If South Korea's plans work out, countries around the world will have a template for making AI a basic digital service. Any failure will prove just as valuable as a lesson. </p> </article> ]]>