AI in parliaments: progress for many, a widening gap between them
The IPU report covers 118 parliaments and chambers from 89 countries, plus two supranational parliaments. AI already helps with daily work, yet the digital divide grows as leading institutions advance faster and governance struggles to keep pace.
The Inter-Parliamentary Union’s World e-Parliament Report 2026 describes a paradox: parliaments are becoming more digitally capable while the distance between the most and least advanced institutions grows. Access to technology is only part of the challenge; institutions also need effective procedures, skills and democratic oversight. [1][2]
Evidence from 89 countries
Conducted between December 2025 and February 2026, the survey covers 118 parliaments or chambers from 89 countries, plus the European Parliament and the Pan-African Parliament. It did not survey 118 countries. Findings rely on institutional self-reporting, with wealthier countries overrepresented. The Digital Maturity Index (DMI) uses a differently consolidated subset; 95 parliaments participated in both the 2024 and 2026 DMI surveys. Changing participants and the revised DMI version limit simple comparisons over time. These findings are not a complete census of the world’s parliaments. [2] (pp. 29 and 84)
Where AI already works—and what changes
88 per cent of responding parliaments report AI use in some form, but only 48 per cent report formal institutional adoption. 78 per cent report informal staff use; this does not mean that 78 per cent of all employees use AI. [2] (pp. 74–76)
Table 42 distinguishes production deployments from pilots. Its production figures are shares of responding parliaments: [2] (p. 77)
- Transcription of proceedings: 39 per cent. In Barbados, AI helped clear decades of backlogged records within months, shifting staff work from transcription to editorial verification. Fiji used existing Microsoft 365 dictation tools to reduce costs and excessive manual workloads. The report describes records reaching citizens in weeks rather than months. [2] (p. 27)
- Translation: 30 per cent; document summaries: 24 per cent. These applications support multilingual access and processing large document volumes. Adoption rates alone establish neither better translation quality nor better political decisions.
- Legislative research and analysis: 14 per cent; enterprise search and knowledge management: 11 per cent. Chile’s ArchileBot provides conversational access to parliamentary archives. Sharing its architecture and validation protocols also illustrates how cooperation can make existing solutions accessible to other institutions. [2] (p. 32)
- Citizen enquiry chatbots: 10 per cent; accessibility improvements: 15 per cent. These applications can broaden access to parliamentary information, but their deployment rates do not demonstrate democratic benefits across all institutions.
The report therefore documents practical improvements and institutional experience, rather than controlled impact evaluations of every application. Human verification remains essential: France’s National Assembly restricts external cloud AI to public information and requires human review of AI-assisted work. [2] (p. 27)
What supports the finding of a widening divide?
Broad progress since 2024: Cloud adoption rose from 68 to 83 per cent, and the share reporting more-than-adequate internet connectivity from 35 to 50 per cent. AI use in any form increased from 28 to 88 per cent. These are improvements in individual indicators, rather than a uniform increase in digital maturity. [2] (p. 21, Table 1)
A growing distance nevertheless: The DMI chapter explicitly reports a modest improvement in median digital maturity since 2024, alongside faster progress at the frontier. Crucially, more than half of the parliaments whose rankings fell nevertheless improved their absolute capability scores. Here, a lower rank reflects slower progress than peers rather than technological decline. This observation supports the report’s central conclusion; changes in rankings alone would not establish a wider absolute gap. [2] (pp. 29 and 31–33)
Current differences also concern AI oversight. All responding parliaments in the highest DMI decile conduct legal and risk assessments; in the lowest, 92 per cent report none of the compliance activities surveyed. Overall, 46 per cent have conducted none of the five measured activities. These are cross-sectional findings for 2026, not separate measurements of change since 2024. [2] (pp. 77–78)
Why the divide could widen further
Further widening is a reasoned risk, not a certain forecast. AI enters everyday work through existing office software while oversight, accountable processes and rules still need to be established. Only 24 per cent report formal AI policy documents. In many institutions, technological availability is therefore advancing faster than the mechanisms needed to govern its use responsibly. [2] (pp. 20 and 75)
Institutions with skilled staff, well-managed data and effective oversight can put new systems to useful work faster. Resource-constrained institutions must build those foundations while technology keeps changing. This suggests that advantages and disadvantages may reinforce one another, affecting parliamentary scrutiny and public access to information.
The IPU’s response is governance before procurement, process redesign before AI integration, and data as a strategic foundation. Clear responsibilities, human oversight, documented risk assessments, training and inter-parliamentary cooperation are central. The digital divide is consequently also a question of democratic institutional capacity. [1][2] (pp. 18–19 and 32–33)
Sources
This indication by Martin Ebner is licensed under
CC BY 4.0
Comments 0
Comments are moderated. New comments appear after review by the editors.
No comments yet.