OIA Analysis – Government AI Transparency: Fragmented, Delayed, and Inconsistent Across Agencies

Primary request: https://fyi.org.nz/request/33964

DIA response:

Summary

This request sought the raw dataset from the 2025 cross-agency AI use survey coordinated by the Department of Internal Affairs.

The request was refused under section 18(d) on the basis that the information will soon be publicly available (expected May 2026).

Key Finding

The refusal confirms that:

• the dataset exists
• it is releasable in principle
• access is being delayed pending publication

This represents a form of deferred transparency, rather than immediate disclosure.

Related OIAs (AI use across government)

This request sits within a growing body of OIAs examining government use of artificial intelligence:

• ACC – Comprehensive AI inventory (tools, contracts, datasets)
https://fyi.org.nz/request/32559

• NZ Police – AI use in video analytics and surveillance
https://fyi.org.nz/request/32947

• Ministry of Health – AI tools and software list
https://fyi.org.nz/request/32435

• Ombudsman – Application of the OIA to AI systems
https://fyi.org.nz/request/33209

Emerging Pattern

Across these requests, a consistent pattern is visible:

1. Macro-level transparency (strong)

• Cross-agency survey coordinated by DIA
• Aggregated results published
• Central AI frameworks and strategy documents available

2. Agency-level detail (inconsistent)

• Some agencies provide limited tool lists (MoH)
• Others delay or do not respond fully (ACC)
• Operational agencies restrict detail (Police)

3. Dataset vs disclosure gap

• Survey data exists centrally
• Agency-specific inventories are fragmented
• Costs, contracts, and deployment details are rarely fully disclosed

4. Delayed release mechanisms

This request demonstrates a key pattern:

> information is withheld temporarily under “soon to be published” grounds

This limits:

• immediate scrutiny
• independent analysis
• cross-agency comparison

Structural Issues Identified

Fragmentation

AI use is not disclosed in a consistent, standardised way across agencies.

Timeliness gaps

Significant delays are evident in high-impact agencies (e.g. ACC).

Lack of search transparency

None of the identified OIAs provide:

• systems searched
• keyword methodology
• verification of completeness

This raises questions about whether responses are systematic or selective.

Governance inconsistency

While DIA/GCDO coordinates cross-government AI activity:

• disclosure practices vary significantly between agencies
• no uniform standard exists for releasing AI-related information

Overall Assessment

Based on available evidence:

> New Zealand government AI transparency is fragmented and partial

• strong at the policy and coordination level
• inconsistent at the operational and agency level

Why this matters

Artificial intelligence is now embedded across:

• service delivery
• decision-making systems
• data analysis and surveillance

Public accountability requires:

• visibility of tools and systems
• understanding of risks and governance
• access to underlying data where appropriate

Next Steps

This request remains active in substance pending publication.

Key follow-up questions will include:

• whether the released dataset includes agency-level detail
• whether raw data is complete or filtered
• how consistent AI use reporting is across agencies

Broader Context

This request forms part of an emerging evidence base examining:

• how government uses AI
• how that use is recorded and disclosed
• whether transparency keeps pace with deployment