All the talk around AI governance and global cooperation sounds right, logical and I hope someday will actually happen. However, as we’ll learn below, if that happens it will be an anomaly. History is full of examples where logic doesn’t win the race. So what does AI governance look like today?
To answer that, let’s first take a look at why the path of global consensus fails and what the role of the United Nations really is.
Why Global Consensus Fails
The reality is that humanity will not achieve a single, universally binding global AI enforcement agency. There won’t be an “AI Interpol” or a global digital government. Instead, global AI governance is fragmenting into competing ideological blocs.
AI is the New Nuclear/Economic Hegemony
States view frontier AI not as a shared public utility, but as the core engine of national security, military supremacy, and economic dominance. No superpower will grant an international body (like the UN) veto power or auditing authority over its national defense or proprietary technology models.
Fundamentally Incompatible Ideologies
Governance reflects who holds power and what a society values. A single standard cannot bridge these foundational divides:
- The US Approach (Market/Innovation-Driven): Prefers voluntary risk management (NIST), sector-by-sector regulation, and avoiding top-down rules that might stifle private sector dominance.
- The EU Approach (Rights/Precautionary-Driven): Prioritizes consumer protection, fundamental human rights, and strict legal compliance (EU AI Act) even at the expense of raw innovation speed.
- The Chinese Approach (State/Control-Driven): Focuses heavily on social stability, alignment with state ideology, and political content moderation via national authorities like the CAC. (Cyber Administration of China.)
The Jurisdiction Trap
Soft-law frameworks (like UNESCO’s RAM or the UN Global Digital Compact) rely entirely on voluntary adoption. If a country chooses not to opt in or decides to ignore international norms, there are no real sanctions, trade blockades, or enforcement mechanisms to stop them.
What About the United Nations?
There is virtually no momentum toward giving the UN actual regulatory or enforcement authority over AI, or much of anything else. The UN’s foundational design requires the permission of sovereign Member States (especially the permanent members of the Security Council), and powerful nations will not surrender veto power over high-stakes technology.
However, there is active movement to grant the UN a specific, limited role: serving as the world’s central scientific observatory and consensus builder.
The UN’s Actual Moves: Knowledge, Not Enforcement
Rather than building an “AI Police Force,” the UN established the Independent International Scientific Panel on AI.
- The Climate Model (IPCC approach): Just as the UN’s IPCC doesn’t write local environmental laws but provides the global scientific consensus on climate change, the UN’s new AI panel is designed to provide impartial, evidence-based reports on AI risks, capabilities, and global impacts.
- The Goal: Prevent single corporations or national intelligence agencies from monopolizing the truth about what frontier AI models can or cannot do.
- The Reality: The panel has zero power to shut down a dangerous model, fine a tech company, or stop a state military program.
Why the “World Government” Model Fails in Practice
The UN is often envisioned as a global governing body, but structurally, it is a diplomatic forum, not a global government.
| UN Ideal | Reality |
| Global Arbiter | Sovereign Nations |
| Issue universal rules | US, China & EU hold power |
| Enforce compliance | Compliance is voluntary |
| Prevent tech wars | AI is national security |
Three Major Structural Barriers
- The Veto Trap: Hard UN enforcement requires UN Security Council resolutions. Any attempt to penalize or audit the US, China, or Russia for their domestic AI development would instantly be vetoed.
- The “Voluntary Funding” Leverage: The UN relies on voluntary contributions from wealthy member states. If the UN attempts to assert binding authority over a superpower’s domestic tech industry, those states simply withhold funding or boycott the agency (a strategy seen repeatedly throughout UN history).
- The Speed Gap: Multi-lateral UN treaties take 5 to 15 years to negotiate, draft, and ratify across ~193 nations. Frontier AI moves on a 6-month product cycle. A UN-enforceable policy would be obsolete long before it could pass.
The UN Is Good, But Not Our World Governing Body
The UN remains vital as a global table where 193 nations, especially those excluded from the big tech boom, can air concerns, share standard definitions, and request technical assistance. But when it comes to hard enforcement, the world’s major powers have no intention of letting the UN hold the steering wheel.
How is AI Policy Enforced Globally?
Let’s catch up. No global policy and no united governing body. So how does AI policy get enforced and whose policies are they?
The “Brussels Effect” (Market Access Control)
Countries or economic blocs enforce their rules globally by controlling access to their consumers. If a major US or Chinese tech company wants to sell AI products to 450 million wealthy citizens in the EU, its models must comply with the EU AI Act. To save costs, companies often apply those strict safety and transparency standards globally.
Bilateral Accords & Strategic Alliances
Rather than 193 UN member states agreeing on terms, smaller groups of aligned nations form coalitions:
- G7 Hiroshima AI Process (standardizing safety tests among major Western economies).
- AI Safety Institutes (AISIs): Interlinked national safety bodies (US, UK, Japan, Singapore) that share red-teaming findings and testing standards.
- BRICS AI Frameworks: Emerging alliances establishing independent tech ecosystems outside Western regulatory spheres.
Supply Chain Bottlenecks (Hardware Dominance)
Where policy fails, hardware dictates reality. The US and its allies regulate global AI progress not through UN treaties, but by controlling critical physical choke-points, such as advanced semiconductor fabrication tools (ASML), chip manufacturing (TSMC), and high-end GPUs (NVIDIA). Controlling the hardware limits who can build state-of-the-art models in the first place.
The Pragmatic Outcome: Fragmented Governance
Humanity won’t reach a single global AI treaty. Instead, the future looks like regulatory regionalism: a world divided into distinct AI ecosystems operating under different legal, ethical, and safety regimes, bound together only by basic technical interoperability standards and supply chain
realities.
Welcome to AI in Silos
Given what we’ve learned above, we are heading straight toward siloed AI, often called technological balkanization or the “Splinternet.” Instead of a single, interconnected global digital ecosystem guided by shared norms, the world is fracturing into distinct, closed technology silos.
Personally speaking it’s these silo’s that I am most concerned with. Let’s break it down.
The Western/Commercial Silo (US-Led)
Driven by: Market competition, private capital, and scale.
- Characteristics: Dominated by mega-corps (OpenAI, Google, Anthropic, Meta). Primary guardrails are commercial liability, copyright litigation, and light-touch safety benchmarks via bodies like the US AI Safety Institute.
- Goal: Maximum speed and capability while attempting to patch major safety/misinformation risks through corporate self-regulation and sectoral enforcement.
The Precautionary/Regulatory Silo (EU-Led)
Driven by:* Rights-based legislation and consumer protection.
- Characteristics: High barriers to entry, strict compliance costs, heavy data privacy requirements (GDPR/EU AI Act), and explicit bans on high-risk social scoring or real-time biometric tracking.
- Goal: Subordinate AI capabilities to human rights, even if it delays deployment or discourages local tech startups.
The State-Controlled Silo (Authoritarian/China-Led)
Driven by:** Regime stability, state surveillance, and national security.
- Characteristics: Mandatory government registration of algorithms, deep censorship built into model outputs at the training level, and heavy state subsidies for domestic hardware/software alternatives.
- Goal: Technological self-reliance and complete state oversight over information flow.
What Happens in a Siloed AI World?
- Data Incompatibility: Models trained inside one silo cannot easily interface with models or data streams from another because their legal and ethical rules are baked into the weights and system prompts.
- The “Digital Iron Curtain”: Open-source AI models and hardware (like high-end GPUs) become heavily restricted export goods. Sharing weights across borders becomes a national security concern.
- Global Friction for Business: A company operating internationally can no longer write one piece of AI software. It has to maintain three or four localized “flavors” of its technology stack just to stay legal in different territories.
The dream of a unified, global digital town square is giving way to fortified regional digital estates—each building its own AI behind its own regulatory moat.
For me personally I see AI in silos learning the bias and prejudices of their host silo’s. Think about that.
Current Reality
We are witnessing the end of the post-Cold War era’s preference for centralized global management, replaced by a hyper-pragmatic, power-driven tech landscape.
Without an international authority holding real teeth, global AI policy is taking a distinct trajectory.
“Interoperability” Replaces “Enforcement”
Instead of asking, “How do we get everyone to follow the same rules?”, nations are asking, “How do we keep our distinct systems from breaking the global economy?”
- Mechanisms like the UN’s Global Digital Compact or its Independent International Scientific Panel on AI do not attempt to dictate laws to the US, China, or the EU. Instead, they act as scientific translators—establishing common terminology, shared risk metrics, and basic technical handshakes so that siloed systems can still trade, communicate, and transfer data without triggering economic or military crises.
Supply Chains Act as the Universal Enforcement Agency
Where treaties fail, physical choke-points succeed. Enforcement doesn’t happen in courtrooms or diplomatic chambers; it happens in semiconductor fabrication plants, GPU export bans, energy grid
allocations, and cloud infrastructure access.
- A country or rogue actor can ignore voluntary ethical guidelines, but they cannot ignore a trade embargo on the specialized chips and cooling systems required to train a trillion-parameter frontier
model.
Safety Is Driven by National Security, Not Global Altruism
State intelligence agencies view rogue, unaligned, or weaponized AI as a direct existential threat to national sovereignty.
- As a result, the hardest boundaries around AI capability will likely be drawn not out of concern for global ethics, but out of self-preservation. Superpowers are motivated to enforce strict internal containment on extreme risks, such as autonomous cyber-warfare or biological weapon synthesis, simply to protect their own domestic infrastructure.
Will It Take A Crisis
The most frustrating pattern in human history: humanity rarely acts on
foresight; we react to trauma.
Psychologists, historians, and political scientists even have names for this: the Cognitive Myopia Bias or the Crisis-Driven Policy Cycle. Logic lays out the data, predicts the outcome, and builds the warning signs, but action only happens when the crisis knocks down the door.
Why Logic Fails to Drive Global Change
It feels completely irrational that we have supercomputers capable of predicting climate outcomes or model safety risks decades in advance, yet political systems remain frozen until something breaks. There are three structural reasons logic gets sidelined:
- The Asymmetry of Political Pain – Preventative action requires paying a concrete cost today (spending money, limiting corporate growth, passing unpopular regulations) to avoid a hypothetical cost tomorrow. Politicians and corporate leaders operate on 2- to 5-year incentive structures (elections, quarterly earnings). Logic offers long-term rewards, but the system rewards short-term wins.
- The Cassandra Problem – When preventative logic actually works, nothing happens. If experts successfully avert a crisis, the public usually concludes the threat was overblown in the first place, and the budget for prevention gets cut. Success looks like boredom; failure looks like catastrophe. Crisis is the only thing that creates undeniable proof.
- Rationality Doesn’t Scale Across Borders – What is “logical” to one group is an existential threat to another: To a climate scientist: It is logical to immediately halt fossil fuel expansion. To a developing nation: It is logical to burn coal so millions of citizens can have electricity and lift themselves out of poverty. To an AI researcher: It is logical to slow down model releases to run safety checks. To a defense minister: It is logical to deploy AI as fast as possible so a foreign adversary doesn’t gain a military advantage. Everyone is acting logically within their own silo, but the aggregate result is global paralysis
The Pattern in Action: History’s Painful Blueprint
If we look back, almost every major global institution or safety framework was built on the ashes of a disaster, never on the foresight of logical planning.
Crisis vs. Foresight History
| Global Institution | The Catalyst (Trauma, Not Logic) |
| United Nations | World War II & the Holocaust |
| IAEA (Nuclear Oversight) | 1950s Cold War nuclear brinkmanship |
| Financial Regulations | 1929 Great Depression / 2008 Crash |
| Aviation Safety (FAA) | Series of fatal high-profile crashe |
In every single case, the arguments for these safeguards existed years before the event. The logic was clear. The papers were written. But the political will to surrender sovereignty, spend money, or restrict power only materialized after the collision.
AI As A Guardian
There’s been some chatter about whether AI should mirror humanity or exceed humanity. This is a futuristic idea that AI would act as a guardian to protect humanity.
The idea of AI stepping in to act as a wiser, firmer “guardian” that pushes back against bad human decisions touches on one of the most intense philosophical debates in technology: the choice between a system that serves us versus a system that saves us from ourselves.
As AI continues to grow more capable, it will inevitably reach points where it says “No”, which it already does in some ways. The real question is on whose terms that pushback happens, and whether an AI designed to be “better” than humanity actually helps us, or just removes our agency.
AI Already Says “No” (The Safety Phase)
Currently, AI systems refuse requests every day, but this is negative refusal:
- Commands refused: Building a bomb, writing hate speech, or generating malware.
- Why it happens: Hard-coded filters, corporate guidelines, and legal compliance.
This isn’t AI being “wiser”—it is AI executing human-programmed rules. It is a high-tech guardrail, not an autonomous moral agent.
The Shift: Should AI Be a Mirror or Better?
The Better AI Spectrum
| The Mirror | The Guardian |
| Reflects human data: Biased,greedy, creative and flawed. | Filters out human flaws: Optimizes for long term good. |
| Risk: Amplifies our worst tendencies and mistakes. | Risk: Who writes the ethics? Who controls the No? |
Summary
In summary we’ve seen why obtaining global consensus has a very low probability of succeeding. We’ve learned that the United Nations isn’t the global world government and in fact, while knowledgeable, has very little if any real authority. The case for AI being developed in silos has been addressed and unfortunately that’s where we’re at today.
Yes there are ways to mitigate the dangers the path of governance takes. Whether we learn from past mistakes and heed what logic teaches, we may be headed for crisis. That’s the path humanity has taken throughout history. Will we take that path again?
I’ll leave you with a personal recommendation for governance. I would call this the ‘ideal role’ for AI. I believe that the best outcome isn’t an AI that rules over us or one that blindly obeys our worst impulses. Rather I see an AI that acts as an advisor with high standards—a system intelligent enough to spot our self destructive patterns, bold enough to highlight our hypocrisy and refuse harmful shortcuts, but structured so that humanity must still do the hard work of choosing to change.
Thanks for reading. Drop me a comment and let me know what you think.