# UN Independent International Scientific Panel on AI: evidence note

Source: *Preliminary Report of the Independent International Scientific Panel
on AI: Evidence-based assessment of opportunities, risks and impacts of
artificial intelligence*, July 2026.

Direct report:
https://www.un.org/independent-international-scientific-panel-ai/sites/default/files/2026-07/en_Preliminary%20Report_.pdf

## Status and limits

- The Panel was established by UN General Assembly resolution 79/325.
- Members serve in personal capacities and the assessment reflects broad, not
  necessarily unanimous, scientific agreement.
- The report is policy-relevant but non-prescriptive. It does not state an
  official policy position of the United Nations or Member States.
- It is preliminary and intended to be updated as evidence develops.

## Findings relevant to Unit Cost Dominance

### Concentration and fiscal exposure, pages 16-18

- Advanced chips, compute, cloud provision and frontier-model access are highly
  concentrated across a small number of firms and countries.
- The Panel says this concentration can generate significant rents.
- It identifies a fiscal risk if production shifts from labour to concentrated
  capital in countries whose revenue systems rely heavily on labour taxation.
- It also identifies regulatory-capture, accountability and dependency risks.

### Mixed labour evidence, pages 18 and 33

- The report cites an approximately 15% relative employment decline among
  United States workers aged 22 to 25 in AI-exposed occupations.
- It contrasts this with Danish evidence showing near-zero effects on
  employment, hours or wages.
- The cross-country difference is evidence that institutions influence the
  effects of the same technology.
- The Panel warns that layoffs attributed publicly to AI do not by themselves
  establish clean causality.

### Distribution is the unresolved core, pages 33-34

- Task-level productivity evidence is positive for some well-defined tasks,
  but micro-level gains do not automatically become economy-wide gains.
- Complementary investment in data, workflows, skills and organisational
  redesign affects the speed and scale of realised outcomes.
- The Panel treats the distribution of surplus between workers, firms,
  countries and capital owners as the central unresolved economic question.
- It says weak current labour effects do not rule out larger later effects.

### Forecast uncertainty, pages 33-34

- Published macroeconomic forecasts span a very wide range.
- The report includes a cited full-automation scenario in which output rises
  sharply while real wages and labour share collapse after very high task
  automation. This is a modelled scenario, not the Panel's forecast.
- The report's own conclusion is uncertainty conditioned by capability,
  adoption, bottlenecks, substitution assumptions and institutional response.

### Measurement, page 34

- National statistical systems need better, privacy-preserving access to data
  that can measure AI's effects on productivity, jobs, wages, trade, firm
  performance and distribution.
- The Panel specifically raises questions about ownership of the AI stack,
  surplus capture and the resilience of labour-income tax systems.

## Appropriate use on the Solutions site

The report supports:

1. treating severe labour displacement as a contingency worth preparing for;
2. publishing an economic-agency dashboard before irreversible spending;
3. monitoring labour share and the shift in taxable income between labour and
   capital;
4. competition, portability and public-capacity policies aimed at concentrated
   rents and dependency;
5. bounded-regret measures whose value does not depend on proving the strongest
   form of the thesis today.

It does not support claims that:

1. mass unemployment is already established;
2. the wage-demand circuit will inevitably collapse;
3. any particular tax, citizen dividend or public-equity rule is endorsed by
   the Panel;
4. the cited full-automation scenario is a forecast or consensus outcome;
5. labour effects will be uniform across countries.