It is tempting to believe that artificial intelligence, this shimmering promise of progress, could one day be governed by a single, coherent rulebook. Having spent over a decade working on international standards in communication technology, I know patience is required. However, standing amid the noise of 2025, one realizes that the current approach is more like a patchwork quilt stretched too thin. Each nation has stitched its own piece, and some stitches, frankly, are already fraying. Regulatory fragmentation produces the same outcome as regulatory absence, uncertainty masquerading as structure.
The Chorus of Laws and the Silence Between Them
In Brussels, the EU AI Act (effective February 2, 2025) now hums like a cathedral organ performing its dissertation on grandeur, beautifully structured, but somehow exhausting. France, Germany, Italy, and the Netherlands each interpret its music differently. The French, with characteristic precision, add their own harmonies: laws against non-consensual Deepfakes and virtual influencers who forget to label themselves as illusions. Germany, methodical as ever, drafted the AI Market Surveillance Act to decide who polices the machines. Italy, meanwhile, seems caught between optimism and fatigue, its Data Protection Authority (Garante) has already fined OpenAI €15 million (December 2024) and Luka Inc. €5 million (May 2025), arguing both ignored transparency duties under the GDPR
The Dutch reflect quietly on the damage done: after their childcare benefits scandal (toeslagenaffaire), algorithmic discrimination is not an abstraction but a national wound. One can almost feel the chill in Den Hague whenever someone mentions “automated decision-making.”
The United States: Fifty Shades of Regulation
Across the Atlantic, uniformity dissolves entirely. The United States has no federal AI law. Instead, it’s a constellation of state initiatives, Colorado’s Algorithmic Discrimination Act (effective 2026), California’s data-disclosure rule for AI developers, and the TAKE IT DOWN Act protecting victims of AI-generated intimate imagery.
Then came President Trump’s 2025 Executive Order, promising to “remove barriers to American leadership.” It rescinded the previous administration’s attempt at “safe and trustworthy AI.” The result? A kind of regulatory vacuum with patriotic flair, encouraging innovation, yes, but leaving compliance officers and legal industry users of AI software with heartburn.
Mediators observing disputes in this space may notice the irony: a startup fined in Italy might find sanctuary in Texas. The same algorithm judged “discriminatory” under the EU AI Act could be deemed “innovative” in Miami.
Asia’s Soft Hands, Firm Shadows
Further east, Singapore, Hong Kong, and Australia take what might be called a “soft-law” path, not weak, but supple. Singapore’s Model AI Governance Framework for Generative AI (2024) reads almost like a set of moral reflections rather than statutes. It speaks of fairness, ethics, accountability, and transparency, the FEAT principles guiding its financial institutions. In Hong Kong, regulators prefer ethical “guidance” to binding law, though even there, the Privacy Commissioner’s 2024 document on responsible AI sounds more like scripture than suggestion. And in Australia, “guardrails”, voluntary today, perhaps mandatory tomorrow, keep developers from straying too close to the edge. China, however, sits in a different register. Its Cyberspace Administration enforces multiple overlapping rules on generative AI, Deepfakes, and algorithmic labeling.
The Middle Path (or No Path at All)
South Africa and Canada, in their own ways, represent the middle ground. South Africa’s National AI Policy Framework (2024) aspires to inclusivity and ethical reflection, still awaiting legal teeth and the ability to enforce. Canada’s attempt to legislate, the Artificial Intelligence and Data Act, died when Parliament was prorogued in January 2025 leaving a void filled mostly by voluntary codes and provincial privacy laws. To anyone accustomed to a stable legal scaffolding, this is dizzying.
The Maze of Compliance
For mediators and arbitrators, AI regulation now forms a kind of labyrinth, not impossible, but easily misread. There are repeating corridors: data protection, anti-discrimination, medical safety, and financial transparency. Yet each corridor opens onto a different jurisdictional tone. The European Union treats AI as potentially dangerous medicine, beneficial if the dosage is correct. The United States treats it more like caffeine: self-regulated, commercially essential, occasionally addictive. Asia experiments with moral frameworks instead of punishments, while Africa, cautious but hopeful, insists on inclusivity and sovereignty over data.
It’s not just the laws themselves but the temporal instability, statutes delayed, frameworks “under consultation,” fines under appeal, that complicates compliance. Consider the EU AI Act’s tiered risk categories: “high-risk,” “limited-risk,” “unacceptable.” Okay, what if an AI shoe salesperson sells the wrong shoes and causes hip damage? Does that fall under high-risk like self-driving cars?
When Law Becomes Language
There’s something almost elegant and dysfunctional about this. The words “trustworthy AI,” “accountable systems,” “human oversight”, all noble, appear in nearly every document from Sydney to Seoul. Yet they mean different things depending on who pronounces them.
Someone once told me that international AI compliance feels “like translating Shakespeare through six dialects while the play is still being written.” Perhaps that’s why many professionals, even in law, approach the subject with quiet anxiety. We sense that regulation is chasing a train that has already left the station, accelerating faster than any lawyer can draft.
Toward Ethics of Navigation
For those of us in dispute resolution, the lesson is humility. We cannot expect uniformity; instead, we must help parties find interpretive bridges between their systems, between the European appetite for oversight and the American instinct for autonomy.
Mediators and arbitrators may need to ask not only what the law says but what the law means in its cultural setting. Compliance, after all, is not just about documents, it’s about behavior, incentives, and narrative.
The fragmentation of AI regulation, then, is not merely a technical challenge; it’s a mirror reflecting our global disunity about responsibility itself. In that mirror, some see risk, others opportunity, and a few, perhaps foolishly, see beauty.
Thanksgiving may be a challenging holiday for parents and children after divorce. Complicated emotions and stress can arise as families navigate new arrangements and family time apart. Many parents “alternate...
By Michael AuritGuest blogger, Carlo Mosca, is a commercial lawyer specializing in international business. He’s been practicing as a mediator, and a trainer in ADR since the mid ‘90s (www.carlomosca.it). He is...
By Dan SimonFor more hilarious law cartoons by the fabulous Charles Fincher, click here. Yesterday, I had the distinct pleasure of speaking to, and then participating in a mock mediation with, Lisa Klerman's...
By Victoria Pynchon
