The Trilemma at the Heart of Global AI Policy

NewsOnScale Staff

June 24, 2026

There is a reason AI regulation feels perpetually unfinished. Governments announce frameworks, industries publish compliance pledges, and international bodies convene summits — yet the rules never quite cohere into something enforceable or consistent. A working paper out of Harvard Kennedy School puts a name to the underlying problem: it's a trilemma, and no country has escaped it.

The argument, in brief, is that three goals dominate the global AI policy conversation — protecting human rights, fostering technological innovation, and preserving national sovereignty over data and infrastructure — and that pursuing any two of them fully tends to compromise the third. This isn't a political failure so much as a structural constraint. It's a framework worth taking seriously, especially for anyone watching how platform power and civic accountability actually play out in practice.

## What the Trilemma Actually Means

Consider the European Union's approach. The EU AI Act prioritizes rights-based protections: transparency requirements, prohibitions on certain biometric surveillance applications, and mandatory human oversight in high-risk domains. Critics, including major technology companies and some member state governments, argue that this comes at the cost of competitiveness — that stringent premarket requirements will push AI development toward jurisdictions with lighter regulatory touch. The EU, in other words, leaned toward rights and sovereignty, and accepted a potential innovation penalty.

The United States has historically leaned the other direction. Federal AI policy has largely been sector-specific and voluntary, with innovation explicitly framed as a strategic national interest. The result is a system where accountability is diffuse and often arrives after harm has already been documented — in hiring algorithms, in predictive policing tools, in content moderation systems that affect political speech at scale.

China represents the third configuration: sovereignty and state-directed innovation, with rights protections substantially subordinated to both. Its regulatory architecture for AI is extensive but oriented toward control rather than protection of individuals from state or corporate actors.

None of these models is internally incoherent. Each reflects a genuine prioritization. The problem is that when these systems interact — when an AI product built under one regime is deployed in another — the gaps become exploitation surfaces. A surveillance technology developed under weak rights frameworks doesn't stay in its country of origin. A data governance standard built for domestic sovereignty doesn't automatically extend to users across borders.

## Why This Matters Beyond Academia

For journalists and civil society organizations covering platform accountability, the trilemma framing is practically useful because it reframes a common frustration. When a regulatory body fails to act on a documented harm, it's tempting to attribute that to captured agencies or industry lobbying — and sometimes that's accurate. But often the failure is structural: the agency is operating inside a policy framework that has already traded away the enforcement tool that would be relevant.

This is especially visible in the political technology space. AI systems that influence voter targeting, content amplification, and political advertising exist in a regulatory gray zone in most democracies. They are too politically sensitive to regulate aggressively under innovation-forward frameworks, too commercially entrenched to be unwound by rights-based challenges, and too globally distributed to be corralled by any single nation's sovereignty claims.

## The Accountability Gap Is the Story

What the Harvard framework ultimately surfaces is that the absence of a unified global approach to AI governance isn't an oversight waiting to be corrected — it's the predictable output of competing national interests that have never been fully reconciled in public.

Citizens in most countries have not been asked which of the three values they'd prefer their government to prioritize. That conversation has happened largely inside trade ministries, national security councils, and industry working groups. The trilemma is real, but the choice about how to navigate it remains, in most democracies, unmade in any accountable public sense.

That is the story. Not that regulation is complicated — everyone already knows that — but that the complication is being managed out of public view, and the values being traded away are ones that affect everyone who interacts with an AI system, which is now nearly everyone.