Every powerful technology begins as a promise. Then it becomes an industry. Then it becomes a governance problem.
Artificial intelligence has reached that third stage.
For years, AI was discussed as a tool of efficiency: better search, better translation, better recommendations, better automation, better analytics. Then generative AI arrived and changed the tone of the conversation. Suddenly, AI was not merely helping humans process information; it was producing text, images, code, music, videos, legal drafts, financial analysis, political propaganda and emotional companionship. It moved from the background of digital life to the front stage of society.
That shift has created one of the defining policy questions of the 21st century: how should governments regulate a technology that is evolving faster than law, spreading faster than institutions, and becoming more powerful than many states can fully understand?
The answer is no longer technical. It is political.
AI regulation is becoming a contest between innovation and control. Governments want to encourage AI because it promises productivity, competitiveness, defence capability, scientific discovery and economic growth. At the same time, they fear AI because it can deepen surveillance, distort elections, automate discrimination, replace jobs, manipulate citizens, weaken privacy and concentrate power in the hands of a few companies or states.
This is why the debate around AI regulation is not simply about technology. It is about power.
The world is now trying to decide whether AI should be governed mainly by markets, by states, by courts, by technical standards, by global treaties or by the companies building the models. Each answer creates a different future.
The Speed Problem
The first problem with AI regulation is speed.
Traditional law moves slowly. It studies harms, builds consensus, drafts rules, debates amendments, passes legislation, creates enforcement bodies and then waits for courts to interpret disputes. AI does not move like that. It improves in months, sometimes weeks. Models become more capable before regulators have fully understood the previous generation.
This creates a permanent gap between innovation and governance.
If governments regulate too early, they may lock in rules for a technology they do not yet understand. If they regulate too late, harms may spread before institutions can respond. If they regulate too strictly, they may kill domestic innovation. If they regulate too softly, they may surrender public interest to private power.
That is the central tension of AI governance: law must be stable, but AI is unstable.
The European Union has tried to solve this through a risk-based framework. Its AI Act entered into force on 1 August 2024, with phased application: prohibited AI practices and AI literacy obligations began applying from 2 February 2025, governance rules and obligations for general-purpose AI models became applicable from 2 August 2025, and the wider framework is moving through staged implementation.
This model says AI should not be regulated as one single thing. A chatbot, a medical diagnosis tool, a biometric surveillance system, a credit-scoring model and an AI weapon system do not carry the same level of risk. The law must therefore classify uses, not merely technologies.
That is a sensible approach. But even the EU’s model faces a hard question: can detailed legislation keep pace with a technology that constantly changes its own capabilities?
The Three Models of AI Governance
Across the world, three broad models of AI regulation are emerging.
The first is the rights-based model, led most visibly by Europe. It treats AI as a technology that must be disciplined by human rights, transparency, accountability and democratic oversight. The EU’s AI Act is built around risk categories, obligations for high-risk systems, restrictions on unacceptable uses and rules for general-purpose AI models.
The second is the innovation-first model, associated strongly with the United States. The US has relied heavily on voluntary frameworks, sectoral rules and technical risk-management standards rather than a single comprehensive AI law. The NIST AI Risk Management Framework, released in 2023, is voluntary and designed to help organisations manage risks to individuals, organisations and society while incorporating trustworthiness considerations into AI design and use.
The third is the state-control model, most closely associated with authoritarian digital governance. In this model, the state sees AI not only as an economic asset but as an instrument of social stability, information control and national security. Regulation is not only about protecting citizens from private harm; it is also about protecting the state from uncontrolled information flows.
Most countries will not fit neatly into one category. They will borrow from all three. India, for example, cannot simply copy Europe’s strict regulatory architecture, America’s market-led experimentation or China’s state-control approach. India’s challenge is different: it must encourage innovation, expand inclusion, protect citizens, build domestic capacity and avoid becoming merely a user-market for foreign AI systems.
India’s governance direction has increasingly focused on safe, inclusive and responsible AI adoption while also recognising the strategic need to democratise AI benefits and build domestic capacity under the IndiaAI Mission.
That balance is difficult. But for India, it is unavoidable.
Why Governments Fear AI
AI regulation is not driven by abstract ethical anxiety. It is driven by concrete fears.
The first fear is misinformation. Generative AI can produce persuasive fake content at scale. Earlier, propaganda required writers, designers, video editors and distribution networks. Now, one actor can generate thousands of messages, images or videos cheaply. In election seasons, this can create confusion faster than fact-checking institutions can respond.
The second fear is deepfakes. AI-generated audio and video can damage reputations, create political panic, enable fraud and violate personal dignity. The danger is not only that fake content will be believed. The deeper danger is that real content will also become deniable. When everything can be fake, truth itself becomes easier to attack.
The third fear is algorithmic discrimination. AI systems used in hiring, lending, policing, education, insurance or welfare delivery can reproduce bias hidden inside data. If a human discriminates, there is at least a visible decision-maker. If an algorithm discriminates, the harm may be buried inside a model, training dataset or automated workflow.
The fourth fear is job displacement. AI may not replace all workers, but it can change the bargaining power of workers. It can compress entry-level roles, automate routine cognitive tasks and make companies less dependent on human labour in some functions. The social impact will not be evenly distributed.
The fifth fear is national security. AI can strengthen cyberattacks, accelerate military targeting, improve surveillance and support autonomous weapons. Once AI enters defence systems, regulation becomes harder because states do not easily disclose strategic capabilities.
The sixth fear is concentration of power. Advanced AI requires data, talent, compute infrastructure and capital. These resources are concentrated among a small number of large technology companies and powerful countries. The OECD notes that AI creates opportunities in healthcare, productivity and scientific progress, but also risks related to disinformation, data insecurity and copyright infringement; it also stresses that AI knows no borders.
This is why AI regulation is also a competition policy issue. The question is not only whether AI is safe. The question is: who owns the infrastructure of intelligence?
Why Innovators Fear Regulation
But the other side of the argument is equally important.
Innovators fear that excessive regulation could slow down experimentation, increase compliance costs, protect incumbents and push smaller companies out of the market. If every AI startup is forced to maintain expensive legal teams, audit systems, documentation layers and approval processes, only the largest firms will survive.
This creates an irony: regulation designed to control Big Tech may end up strengthening Big Tech.
Large technology companies can absorb compliance costs. Startups often cannot. Large firms can hire policy teams, lawyers, safety researchers and auditors. Small innovators may struggle even to understand the regulatory burden. If governments are not careful, AI regulation may turn into a licensing regime for giants.
This is why the innovation-versus-control debate is not a simple moral debate. It is not that regulators are responsible and innovators are reckless. Nor is it true that innovators are visionary and regulators are enemies of progress. Both sides have legitimate concerns.
The real challenge is designing regulation that targets harmful uses without freezing useful experimentation.
A country that bans too much may become safe but irrelevant. A country that permits everything may become innovative but unstable. The best AI governance system must avoid both extremes.
The Global Governance Problem
AI does not respect borders.
A model trained in one country can be deployed in another. A deepfake generated in one jurisdiction can influence politics in another. A chatbot hosted abroad can serve users domestically. A foreign AI model can shape education, news consumption, financial decisions and public debate inside another country.
This creates a governance problem that national laws alone cannot solve.
The OECD AI Principles, first adopted in 2019 and updated in 2024, promote innovative and trustworthy AI that respects human rights and democratic values. They provide flexible guidance for policymakers and AI actors across jurisdictions.
The Council of Europe’s Framework Convention on Artificial Intelligence, opened for signature on 5 September 2024, became the first international legally binding treaty in this field. Its objective is to ensure that AI systems remain consistent with human rights, democracy and rule of law while still allowing technological progress.
These initiatives matter because AI governance cannot remain only domestic. But international coordination will be difficult because countries do not share the same political values.
A democracy worries about misinformation, privacy and rights. An authoritarian state worries about social control and regime stability. A developing country worries about access, affordability and technological dependency. A superpower worries about strategic advantage. A corporation worries about market share, liability and speed.
Everyone says they want “responsible AI.” But they do not all mean the same thing.
The India Challenge
For India, AI regulation must answer five questions at once.
First, how can India protect citizens without suffocating startups?
Second, how can India encourage AI adoption while preventing irresponsible deployment in sensitive sectors?
Third, how can India build domestic AI capacity instead of depending entirely on foreign models?
Fourth, how can India ensure that AI works across Indian languages, social contexts and institutional realities?
Fifth, how can India prevent AI from deepening inequality between those who can use it and those who are used by it?
India’s situation is unique because of scale. A flawed AI system in a small country may harm thousands. A flawed AI system in India can harm millions. If AI enters welfare delivery, education, credit, healthcare, policing or recruitment without safeguards, the damage can become systemic.
But India also cannot afford fear-driven delay. AI can improve governance, agriculture, education, healthcare access, legal research, tax administration, logistics, skilling and public service delivery. For a country with massive administrative burdens, AI is not a luxury. It can become a productivity multiplier.
This is why India needs a calibrated framework: strict where rights and safety are at stake, flexible where experimentation is necessary, and strategic where national capacity is involved.
India should not regulate AI only as a consumer-protection problem. It must regulate AI as a development, sovereignty and competitiveness problem.
The Open-Source Dilemma
One of the hardest questions in AI regulation is open-source AI.
Open-source models allow researchers, startups, students and smaller companies to innovate without depending entirely on closed corporate systems. They reduce concentration of power and encourage transparency. For countries like India, open-source AI can help build domestic capacity and local-language tools.
But open-source models also create regulatory anxiety. Once powerful models are released widely, they can be modified, misused or deployed without central control. Bad actors can use them for cybercrime, disinformation, fraud or harmful automation.
So regulators face another difficult balance.
If they restrict open-source AI too much, they may strengthen closed corporate monopolies. If they allow unlimited release of powerful models, they may lose control over misuse.
The future of AI governance will depend heavily on how this balance is handled. The open-source question is not merely technical. It is democratic. A world where only a few companies control advanced AI is dangerous. A world where anyone can weaponise advanced AI is also dangerous.
Good regulation must preserve openness without ignoring risk.
The Problem of Explainability
Another major challenge is explainability.
Many AI systems do not work like traditional software. In traditional software, a programmer writes instructions and the machine follows them. In advanced machine learning, the system learns patterns from data, and even developers may not fully understand why it produces a particular output.
This creates legal difficulty.
If an AI system rejects a loan application, who explains the decision? If an AI tool wrongly flags a student, who is accountable? If an AI medical system gives a harmful recommendation, who carries liability — the hospital, the software provider, the model developer, the data supplier or the human operator?
Without explainability, accountability becomes weak.
But explainability itself is not simple. Some AI systems can be interpreted reasonably well; others are too complex for easy explanation. Regulators may demand transparency, but technical transparency does not always equal public understanding.
A 200-page model documentation file may satisfy compliance but fail the citizen.
The real goal should be meaningful accountability, not cosmetic transparency. Citizens do not need to understand every mathematical layer of a model. But they do need the right to know when AI is being used, how decisions are reviewed, what data is relevant, how errors can be challenged and who is responsible for harm.
Regulation as Geopolitics
AI regulation is now part of geopolitics.
The country that sets AI rules may influence global standards. The companies that comply first may shape compliance markets. The states that build trusted AI ecosystems may attract investment. The countries that move too slowly may become dependent on foreign platforms.
This is why the EU’s AI Act matters beyond Europe. Even non-European companies may adjust their systems to access the EU market. This is similar to what happened with data protection after the GDPR: one jurisdiction’s rules can influence global corporate behaviour.
But AI regulation may become more contested than data protection because AI is directly linked to military power, economic productivity and ideological influence.
The United States will not want rules that weaken its technology giants. China will not accept governance standards that undermine state control. Europe will try to export rights-based regulation. India will seek strategic autonomy. Smaller developing countries will worry about being rule-takers rather than rule-makers.
The result will be regulatory competition.
AI will not be governed by one global law. It will be governed by overlapping regimes, technical standards, corporate practices, national laws and international principles. The challenge for businesses will be compliance across jurisdictions. The challenge for citizens will be protection across borders. The challenge for governments will be enforcement against systems they may not fully control.
What Good AI Regulation Should Do
Good AI regulation should not be anti-innovation. It should be anti-abuse.
It should not punish every experiment. It should focus on high-impact deployment.
It should not create paperwork for its own sake. It should create accountability where real harm is possible.
It should not treat all AI systems as equally dangerous. It should distinguish between low-risk tools and systems affecting rights, safety, livelihoods and democracy.
It should not protect incumbents. It should keep space open for startups, researchers and public-interest innovation.
It should not be purely national. It should align with global standards where possible while protecting domestic priorities.
Most importantly, AI regulation should be built around one principle: the more power an AI system has over human life, the stronger the accountability must be.
A movie recommendation algorithm does not need the same scrutiny as an AI system used in policing. A grammar assistant does not need the same oversight as an AI system used in welfare eligibility. A customer-service bot does not need the same regulatory burden as an AI system used in hospital triage.
Risk must decide regulation.
The Final Question: Who Controls Intelligence?
At its deepest level, AI regulation is about control over intelligence.
For centuries, intelligence was human, institutional and social. It lived in schools, universities, courts, bureaucracies, companies, laboratories and media organisations. AI changes this structure. It turns intelligence into infrastructure. It allows decision-making, prediction, persuasion and creation to be automated at scale.
That is why the governance stakes are so high.
If AI is controlled only by corporations, public interest may become secondary to profit. If AI is controlled only by states, freedom may become secondary to surveillance. If AI is left entirely open, misuse may become uncontrollable. If AI is overregulated, innovation may move elsewhere.
There is no perfect answer.
But there is a dangerous answer: doing nothing.
A society that refuses to regulate AI will not remain free of regulation. It will simply be regulated by code, platforms and private terms of service. The choice is not between regulation and no regulation. The choice is between democratic regulation, corporate regulation and state control.
That is the real contest.
AI regulation is not a brake on the future. Done well, it is the condition for public trust. And without public trust, even the most powerful technology eventually faces resistance.
The age of artificial intelligence will not be judged only by what machines can do.
It will be judged by whether human societies had the wisdom to govern them before they governed us.
Language note: “twelth” should be written as “twelfth.”
Start writing the twelth article.
I’m taking “twelfth article” as Article 60 from the 49–71 list: “Multipolarity Gives India More Space but Also More Risk.” I’ll frame it as a strategic editorial with current examples, India-first analysis, and SEO components upfront.