The Genie Argument
You cannot put it back.
The only question is what you do next.
Every powerful technology in human history has generated the same fear response: ban it, contain it, wall it off from those who might misuse it. The printing press was censored. The telegraph was nationalised. Nuclear knowledge was classified. The internet was regulated, filtered, and in some countries blocked. None of these responses stopped the technology. They shaped who controlled it, who accessed it, and who was left behind. The response to AI in 2026 is following the same pattern — and the consequence of getting it wrong is larger than any previous version of this error, because the technology itself is larger than any previous version.
AI is not a product. It is not a platform. It is a general-purpose cognitive capability — the first technology in human history that can augment the thinking process itself across every domain simultaneously. When governments respond to this with national silos, incompatible regulatory regimes, export controls on models and chips, and data localisation mandates, they are not making AI safer. They are making it more expensive, less accessible, more unevenly distributed, and more likely to be developed in the dark by actors who are not subject to any governance at all. The fear of misuse is producing the conditions for exactly the misuse that was feared.
"The alternative — a Balkanised regulatory landscape built around isolated national priorities — risks undermining trust, slowing innovation, and amplifying geopolitical friction." — AI Business Review, February 2026
The Fracture Is Already Happening
Not a warning about the future.
A record of the present.
The fragmentation of AI governance is not a theoretical risk that thoughtful policy could prevent. It is already underway — documented, dated, and accelerating. In the twelve months between June 2025 and June 2026, the regulatory landscape moved from imperfect-but-shared to fundamentally incompatible. The EU, the United States, Russia, and China are not building different versions of the same framework. They are building different internets, different AI ecosystems, and different visions of what the technology is for — and the walls between them are rising faster than the dialogue that might keep them from becoming permanent.
Aug 2024 →
Aug 2026
EU AI Act phases in. High-risk AI obligations enforceable from 2 August 2026. Penalties up to €35 million or 7% of global turnover for prohibited practices. Applies to any company serving EU customers, regardless of where it is headquartered — the extraterritorial reach alone creates compliance costs that disproportionately favour large incumbents and disadvantage smaller innovators and the Global South.
Dec 2025
Trump signs Executive Order 14365. US AI policy pivots to "minimal burden" and "sustaining US global dominance." Creates a task force to challenge state AI laws deemed inconsistent with federal policy. Directly incompatible with the EU framework — a company complying with one regime is structurally non-compliant with the other in multiple areas.
Jan 2026
Multiple US state AI laws take effect simultaneously — California's Transparency in Frontier AI Act, Texas's TRAIGA — creating a third, fourth, and fifth incompatible compliance layer within a single country. A company operating across US states and the EU now navigates six or more distinct AI legal frameworks simultaneously.
Mar 2026
Russia's RuNet isolation decree enters force. Regulators gain legal power to reroute all Russian internet traffic in real time and, if Moscow chooses, sever the RuNet from the global internet entirely. A parallel, governable, self-contained digital infrastructure — built specifically so that AI, data, and communication can be managed without reference to any external standard.
Sep 2025 →
ongoing
China exports its Great Firewall as a commercial product. 600 gigabytes of technical blueprints leaked publicly reveal that China's internet filtering infrastructure is being sold as a ready-made kit to Kazakhstan, Pakistan, Ethiopia, and Myanmar — censorship as a service, scaling the silo model to governments that could not build it independently. Each new buyer is a new walled garden that AI cannot cross without permission from the state controlling the wall.
May 2026
US reverses again toward federal oversight and pre-market review of advanced AI models — the third distinct US AI policy position in eighteen months. Policy instability is itself a form of fragmentation: companies cannot build for a governance environment that changes direction faster than development cycles run.
€35M
EU AI Act maximum penalty — or 7% of global turnover, from August 2026
4+
Countries now running China's Great Firewall as a licensed commercial product
6+
Incompatible AI legal regimes a single global company must navigate in 2026
The Self-Made Catastrophe Pattern
We have done this before.
Every time, it was our choice.
The argument this letter makes is not that AI is without risk — it has risks, significant ones, and they deserve serious engagement. The argument is that the history of human catastrophe is almost entirely a history of self-made disasters, not of technologies that destroyed us despite our best efforts. Climate change was not visited upon us by nature — it was produced by the choices of industrial economies that understood the consequences for decades before acting. Trade wars throughout history have never produced the domestic protection they promised — they have produced retaliatory spirals that cost both sides more than the original grievance. Human conflicts, from the Great War to every subsequent iteration, have overwhelmingly originated in misunderstanding, fear, and the decision to build walls rather than the understanding to build bridges.
Trade Wars · Smoot-Hawley 1930
Protect domestic industry. Fragment global trade.
The tariffs were designed to shield American workers from foreign competition. The result: global trade collapsed, unemployment reached depression levels, and the domestic industries the tariffs were meant to protect suffered more than if they had faced the original competition. Letter 40 in this series documented the same pattern repeating in 2025-26. Nothing was learned.
Climate · 1970s–2020s
Understand the mechanism. Delay the response. Pay compound interest.
The science of anthropogenic climate change was established in the 1970s. The response was national energy interests, sovereign exemption claims, and fifty years of frameworks that fragmented responsibility precisely enough to ensure no single actor felt obligated to act at the required scale. The catastrophes that are now arriving were not surprises — they were choices made across five decades of deferral.
The Internet · 1990s–2020s
Open architecture. Then national walls. Then the splinternet.
The internet was built as an open, interoperable, borderless network. It took thirty years to fragment it — through the Great Firewall, Russia's RuNet, the EU's digital sovereignty frameworks, and now China's exported censorship-as-a-product. The splintering of the internet did not make any country safer. It made information less free, innovation less efficient, and the gap between connected and disconnected populations wider.
AI · 2024–2026
The same choice, being made right now, in real time.
The EU builds a compliance fortress. The US pivots three times in eighteen months. Russia isolates. China exports its walls. Global South nations are given a choice between the EU's compliance burden and China's surveillance infrastructure — neither of which was designed with their interests in mind. The splinternet is becoming the split-AI. The question is whether we notice before the walls are too high to dismantle.
Heaven or Hell — The Two Paths
Embrace it thoughtfully.
Or build the catastrophe yourself.
The Embrace Path · What Thoughtful Looks Like
- Shared minimum standards — not identical regulation, but interoperable foundations on safety, transparency, and accountability that allow models to cross borders without creating compliance paralysis
- AI literacy as a governance prerequisite — legislators and regulators who understand what they are regulating, not who fear what they cannot explain
- Global South inclusion in norm-setting — the 6 billion people not in the EU or US have as much at stake as the 1 billion who are, and frameworks designed without them will be rejected or circumvented by them
- Open research infrastructure — publicly funded, openly published AI safety and alignment work that benefits every governance framework rather than being captured by national security classification
- Proportional, risk-based rules — governing the applications that genuinely create danger (autonomous weapons, mass surveillance, critical infrastructure control) without making the entire technology category a compliance obstacle
The Fragment Path · Where We Are Currently Heading
- Six incompatible compliance regimes for a single global company — the cost of which falls on smaller actors and the Global South, not on the incumbents who can afford the lawyers
- National AI champions built behind regulatory walls, optimised for domestic politics rather than global performance, producing the mediocrity that state-protected industries always produce
- Surveillance AI exported without standards — China's Great Firewall-as-a-service is not the only model available to authoritarian governments; it is simply the most visible one
- AI development moving to ungoverned spaces — when compliance costs become prohibitive in regulated jurisdictions, development migrates to jurisdictions with no standards at all, producing exactly the uncontrolled AI the regulations were meant to prevent
- The knowledge gap compounding — as AI capability diverges between nations that embrace and nations that wall off, the gap between them becomes a source of conflict rather than cooperation
84
Internet shutdowns in India in 2024 alone — the open internet already fragmenting
3×
US AI policy direction changes in 18 months — instability is itself a governance failure
6B+
People in the Global South largely absent from the AI governance frameworks being built for them
The Thing That Could Change This
The 2025 WSIS+20 conference reaffirmed the multistakeholder model of internet governance — a framework in which governments, civil society, technical experts, and the private sector share norm-setting authority rather than surrendering it to any single power. Applying this model to AI governance — with real enforcement capacity, not merely ethical guidelines — is the mechanism that could produce interoperable standards without sacrificing legitimate national policy priorities. The G20, under South Africa's 2025 presidency, with UNESCO as a privileged knowledge partner, has the institutional architecture to begin this work. Whether the political will to use it exists is the open question of 2026.
The Honest Complication — Not All Regulation Is Fragmentation
This letter should not be read as an argument against AI regulation. It is an argument against the specific kind of regulation that is proliferating in 2026 — incompatible, nationally siloed, driven by political fear rather than technical understanding, and applied without regard for the compliance costs that fall most heavily on those with the least capacity to bear them. There are legitimate, urgent AI governance questions that deserve rigorous regulatory answers: autonomous weapons systems, AI-enabled mass surveillance, algorithmic discrimination in consequential decisions, the concentration of AI capability in a handful of private companies. The problem is not that governments are trying to govern AI. The problem is that they are governing it in ways that produce fragmentation without producing safety.
The Honest Read
The most important thing to say plainly is that the people driving fragmentation are not villains — they are frightened, and some of their fear is legitimate. AI does pose genuine risks: to employment, to democratic discourse, to the concentration of power, to privacy, and potentially to human oversight of consequential decisions. A legislator who does not understand the technology and fears what they cannot explain is not making an irrational choice when they reach for the regulatory tools available to them. The problem is that those tools, applied without understanding, produce outcomes that are worse than the risks they were meant to address.
The climate parallel is the most instructive because it shows what happens when the response to a genuine, understood risk is systematically fragmented by national interest. The risk was real. The fragmented response to it was also real. And the consequence of choosing fragmentation over cooperation is now arriving as measurable, irreversible damage that no amount of subsequent cooperation can fully undo. The window for a different choice on AI is still open. It will not stay open indefinitely. The walls being built in 2026 — the EU's compliance fortress, the RuNet isolation, the exported Great Firewall, the proliferating state-level US laws — are not yet too high to dismantle. But they are being built faster than the dialogue that could prevent them from becoming permanent.
The NGE View
The verdict.
What We Believe
✓
The genie is out. Every policy decision made from this point is about shaping the outcome, not preventing the technology. Governments that have not yet internalised this are making choices based on a false premise — that sufficiently restrictive regulation can contain AI within national borders in the way it could contain a factory or a mine. It cannot. The technology is information. Information moves.
✓
The fragmentation being produced by fear is more dangerous than the technology that triggered the fear. Six incompatible compliance regimes, national AI silos, exported surveillance infrastructure, and AI development migrating to ungoverned spaces are each individually more harmful than the unregulated AI that the regulations were meant to prevent. This is not a paradox — it is the predictable outcome of governing what you do not understand.
✓
AI literacy is now a prerequisite for governance, not an optional supplement to it. A legislator who cannot explain the difference between a large language model and a rule-based algorithm should not be writing the law that governs either. This is not elitism — it is the same standard we apply to financial regulators, pharmaceutical regulators, and aviation authorities. You must understand what you regulate.
✓
The Global South is the most important and most neglected stakeholder in this entire conversation. Six billion people are having their AI future decided by frameworks designed in Brussels and Washington without their meaningful participation. The countries that receive China's exported Great Firewall are not choosing surveillance — they are accepting the only infrastructure being offered to them. Inclusive governance is not an idealistic aspiration. It is the condition under which any global AI framework can actually hold.
✓
Embrace it thoughtfully — or build the catastrophe yourself. The climate catastrophe was not created by climate. It was created by the decision, made repeatedly over fifty years, to fragment the response to a known risk along lines of national interest. The trade catastrophes of the 1930s and the 2020s were not created by trade. They were created by the decision to protect the familiar at the cost of the functional. The AI catastrophe, if it arrives, will not be created by AI. It will be created by the same human pattern that created every catastrophe before it: misunderstanding, fear, and the choice to build walls instead of the understanding to build bridges.
The window is open. The technology exists. The walls are rising but not yet irreversible. The choice between embrace and fragment is being made right now — in legislative chambers, in executive orders, in regulatory filings, in export control lists, and in the decision of every government that receives a surveillance infrastructure kit and chooses to install it. Heaven — the chance, the opportunity, the positive-sum future that genuinely thoughtful AI governance could produce — is available. It requires only the combination of understanding and courage that has been available in every previous crisis of the same kind, and that has been deployed just often enough in human history to produce the progress this series has documented across forty-six prior letters. Hell is also available. It requires nothing more than what is already happening. The genie is out. The only remaining question is which world we are building for it to inhabit.
NGE · A Futuristic Investment Letter
Long-horizon thinking on capital, technology, and the forces shaping the next decade of wealth creation. Written from first principles. Not consensus. Not noise.
— Pawan Bhatia · NextGen Economics · Bangalore, India