The Rights Gap at the Heart of Global AI Policy

NewsOnScale Staff

August 20, 2026

Spend enough time reading AI governance proposals and a pattern emerges: documents that are long on principle and short on enforcement. Risk frameworks. Transparency guidelines. Voluntary commitments from industry. The language is careful, the timelines are vague, and the penalties — when they exist at all — tend to fall well below the cost of compliance for major platforms.

A recent assessment from Harvard Kennedy School puts a sharper frame on why this keeps happening. The argument is structural: governments pursuing AI policy are simultaneously trying to protect human rights, remain economically competitive, and maintain meaningful regulatory control. Satisfying all three at once, the analysis suggests, is extraordinarily difficult. Prioritize rights protections too aggressively, and investment flows to jurisdictions with lighter rules. Prioritize competitiveness, and rights frameworks become decorative. Cede regulatory control to industry self-governance, and accountability disappears almost entirely.

This isn't a new observation in political economy, but its application to AI governance is unusually acute — and unusually consequential.

## Why the Trilemma Bites Harder Here

Most industries that regulators have grappled with — financial services, pharmaceuticals, telecommunications — move slowly enough that governments can iterate. A harmful drug can be pulled from shelves. A deceptive lending practice can be unwound. AI systems embedded in hiring pipelines, criminal risk scoring, content moderation, and benefits determination move faster, scale wider, and leave less traceable harm.

The people most exposed to that harm — applicants rejected by automated screening, benefits claimants denied by algorithmic review, communities subjected to predictive policing tools — are also typically the least resourced to challenge those decisions. When transparency requirements are weak and appeal mechanisms are absent, the accountability loop simply doesn't close.

That's the human rights dimension that tends to get soft-pedaled in governance conversations dominated by enterprise compliance concerns. The question isn't just whether a company has an AI ethics policy. It's whether a person harmed by that company's AI system has any realistic path to remedy.

## What Voluntary Frameworks Actually Produce

The current global landscape is dominated by frameworks that rely heavily on self-attestation. Companies declare their systems are fair, conduct internal audits, publish responsible AI principles, and largely determine for themselves what counts as acceptable risk. Regulators in most jurisdictions lack the technical staff, legal authority, or political will to challenge those determinations in any systematic way.

This isn't cynicism — it's observable pattern. The EU's AI Act, the most ambitious binding framework to date, contains meaningful provisions but also substantial carve-outs and phase-in periods that soften its near-term impact. In the United States, federal legislative action remains stalled, leaving a patchwork of state-level rules and sector-specific agency guidance that sophisticated actors can navigate around.

The competitive pressure Harvard's analysis describes is real and observable in that stalling. Legislators who might otherwise support stronger enforcement provisions hear from domestic industry that overregulation will cede ground to less scrupulous foreign competitors. Whether or not that argument is empirically sound, it has proven politically effective.

## Accountability Without Conspiracy

None of this requires bad faith on anyone's part to be a serious problem. Regulatory capture doesn't need corrupt actors — it just needs incentive structures that consistently favor incumbent players over diffuse public interests. When the entities being regulated are also the primary sources of technical expertise available to regulators, and when the harms are distributed across millions of individual interactions rather than concentrated in visible events, the conditions for weak oversight are structurally baked in.

What would change that? Independent algorithmic auditing with real enforcement teeth. Meaningful standing for affected individuals to challenge automated decisions. Public funding for technical regulatory capacity that doesn't depend on industry cooperation. These aren't novel ideas — they appear in academic literature, civil society proposals, and occasional legislative drafts. They are consistently absent from the frameworks that actually pass.

The trilemma Harvard identifies is real. But it's worth being precise about which horn of it most governments are actually choosing. The evidence so far suggests it isn't the human rights one.