PUBLIC SECTOR · 8 MIN READ

Public Sector AI Is Facing an Accountability Test.

The opportunity to improve public services is significant, but so is the burden of explaining how systems are selected, used, challenged and governed when decisions affect citizens.

INTAI Research · Published 30 June 2026

Read the Analysis

Adoption Has Crossed a Threshold the Governance Has Not Yet Matched

Government adoption of AI has accelerated sharply. Federal AI use cases in the United States more than doubled year over year in 2025, crossing 3,000, with the steepest increases at NASA, the Department of Health and Human Services, the Department of Veterans Affairs, the Department of Justice and the Department of Energy. The debate inside most public sector organisations has genuinely moved past whether AI belongs in government, toward the harder question of how to deploy it responsibly at the scale adoption has already reached.

That shift matters because the public sector carries an accountability burden private organisations generally do not: decisions made or influenced by government AI systems affect citizens who did not choose the vendor, cannot easily opt out, and are entitled to an explanation when a decision goes against them. Meeting that burden requires more than the technology working correctly; it requires being able to show, after the fact, how a system was selected, how it was tested, and how a citizen can challenge a decision it contributed to.

The Numbers Behind the Accountability Gap

The gap between adoption and governance readiness is not a matter of interpretation, it shows up directly in how public sector leaders describe their own preparedness.

Seventy-seven percent of federal leaders say oversight frameworks for agentic AI are essential. Fewer than a third have actually built one. Only 20% have defined policies for pre-deployment testing of AI systems before they go live, and only 8% have a genuine incident response framework for when something goes wrong after deployment. On transparency specifically, progress has been measurable but still modest: the share of OECD countries reporting either a transparency standard or a public algorithm register rose from 21% to 31%, real movement, but still leaving the clear majority of countries without either mechanism in place.

Central Insight

A 77% versus “fewer than a third” gap between leaders who believe oversight is essential and organisations that have actually built it is not a technology problem. It is a resourcing and prioritisation problem, and it means that, in most public sector organisations today, deployed AI systems are running ahead of the very accountability mechanisms their own leadership already agrees they need.

What Closing the Gap Actually Requires

The 2026 priority for public sector AI governance is less about new use cases and more about building the specific mechanisms this data shows are missing.

That means formalising clear AI frameworks that address data security, ethical use, accessibility and oversight together, rather than as separate initiatives running on different timelines. It means building genuine pre-deployment testing policies and incident response frameworks, given how few organisations currently have either. And it means treating algorithmic transparency, publishing what systems are in use and how they are governed, as core to public trust rather than an optional disclosure exercise, since the organisations succeeding at scaled AI adoption are, by most current accounts, the ones treating transparency and accountability as what makes continued adoption possible, not as a constraint on it.