A few months ago, on the opening day of the CPN-UML’s general convention in Sallaghari, a photograph began circulating online showing a crowd far larger than anyone present remembered. The image, as it turned out, had been generated by Google’s Gemini and stitched onto a real photograph. By the time the correction arrived, the fake image had already done its work. Most people who shared it never saw the clarification.
I bring this up because we have already had our first taste of what artificial intelligence does to public life in Nepal, and it has shown how unprepared we are. Yet when I read Nepal’s newly approved National Artificial Intelligence Policy of 2025, I do not get the sense that the people who wrote it have spent a single afternoon prompting a model, training one, or losing an argument to a chatbot in a college dorm. The document is well-meaning. It is also a document about a technology written largely by people for whom that technology still feels distant.
Let me first be fair to the policy. The Council of Ministers approved it on 11 August 2025, and on paper, it is more ambitious than I expected. It proposes an AI Regulation Council, a National AI Centre under the Ministry of Communications, AI Excellence Centers in every province within five years, regulatory sandboxes for experimentation, data protection and ethical guidelines, and a built-in two-year review cycle. It also speaks of integrating AI into health, agriculture, education, and public administration. None of this is small. For a country that took two decades to meaningfully implement digital signatures, this is, in fact, somewhat surprising.
Revised interest rate corridor system introduced
But a policy is not the same as a plan, and a plan is not the same as a national habit. The document was drafted and endorsed by the previous government, which fell less than a month later in the September Gen Z uprising. It was inherited by an interim administration and now sits before a new coalition formed after the March 2026 vote. Almost none of the people who approved it in Cabinet are still in office. And almost none of the people who will actually live with its consequences were in the room when it was written—no schoolteachers, no district doctors, no farmers’ cooperatives, no 23-year-old programmers, and no journalists who have ever had to verify a viral video at midnight.
This is what should worry us. AI is not a sector like telecom or banking. It behaves more like electricity. It will quietly enter our classrooms, hospitals, courts, banks, ministries, fields, and newsrooms whether we formally prepare for it or not. If the people writing the rules have only encountered AI through alarming headlines and PowerPoint slides, our policies will reflect that limited familiarity. We will get prohibitions where we needed pilots, certifications where we needed competence, and committees where we needed classrooms.
I want to be clear: the risks are real. Misinformation that looks like evidence. Privacy violations disguised as personalization. Automated decisions in hiring, lending, and welfare that quietly inherit biases from the data they are trained on. Real disruption to clerical, creative, and back-office work that millions of Nepalis depend on. None of this should be dismissed. But the answer to risk is not fear. The answer is competence—distributed widely, taught early, and embedded into how citizens, civil servants, and institutions learn to think.
So who should be in the room when this policy moves from paper to practice? Students who already use AI to write essays and learn to code. Teachers who can see both cheating and genuine learning breakthroughs at the same time. Doctors in district hospitals who know exactly where a second pair of algorithmic eyes would help a tired junior radiologist. Farmers who lose crops to diseases that an image classifier could have flagged earlier. Journalists who already spend half their time chasing misinformation. Civil servants who know which files move slowly because they must, and which move slowly because no one has read them. Lawyers who understand that consent buried on page nine of terms of service is not meaningful consent. None of these voices were at the drafting table. All of them must be at the implementation table.
If I had to compress this into actionable priorities, I would name three. First, AI literacy must run from schools to the civil service—not as an elective, but as a basic skill, like English or digital literacy. Teachers must be trained first; we cannot punish students for using a tool we have not taught them to understand. Second, the regulatory sandbox must become a real, functioning space—not just a line in a policy document. Startups, hospitals, banks, agricultural cooperatives, and ministries should be able to test AI systems in controlled environments before deploying them to citizens. Third, regulation must be tiered. A chatbot helping a small business write product descriptions is not the same risk category as an algorithm determining access to public benefits.
There is also a quieter challenge we cannot ignore. Most of the world’s most powerful AI systems are trained primarily on English, with limited Nepali representation. Devanagari is handled inconsistently, and many local languages are barely represented at all. If Nepal does not invest seriously in Nepali-language datasets and tools that work for citizens most comfortable in Doteli, Maithili, Tharu, or Newar, this revolution will arrive as yet another imported system that benefits only a small urban, English-speaking minority. India has already begun investing heavily in its IndiaAI mission. The European Union has its AI Act. Even Bhutan has begun developing a national AI roadmap. Our policy must be matched with funding, compute infrastructure, and people—or it will remain a document, not a direction.
Nepal has historically arrived late to almost every technological shift—industrialization, personal computing, and the smartphone economy. Each time, we have spent years catching up to consequences others had already absorbed. This may be the first moment in our history when we are arriving early enough to actually shape the outcome. It would be a strange kind of failure to once again hand the pen to those most afraid of the page.
The National AI Policy 2025 is a beginning. Whether it becomes a serious national framework or a forgotten file depends on a single question: whose hands shape what happens next? The answer cannot be limited to ministers and committees. It must also include students, teachers, doctors, farmers, programmers, journalists, and civil servants who will live with this technology long after the next government has changed. AI will not wait for our consultations. The least we can do is ensure the right people are in the room.
