# AI Index 2025: the state of AI in politics and public power

Author: Fernando Nieto Lobato
Original publication: 2025-04-23
Spanish original: https://estrategiabyaleph.substack.com/p/estrategia-82-ia-y-poder-publico
English URL: https://elcontemplador.github.io/estrategia-english/essays/082/
Status: Published translation

This is a translation of the original Spanish essay published on 23 April 2025. Its claims, examples and forecasts retain that historical context.

English publication: 2026-09-29

*Archive note: published on 23 April 2025. This analysis preserves the report figures, policy timelines and investment announcements as described in the original article; it is not an update on their subsequent implementation.*

The year 2024 marked a turning point in the relationship between artificial intelligence, politics and global governance. What until recently had been theoretical debates and nascent national strategies became a wave of concrete action: binding regulations, massive public investment and an unprecedented intensification of international cooperation — and competition. **The [Artificial Intelligence Index Report 2025](https://hai-production.s3.amazonaws.com/files/hai_ai_index_report_2025.pdf), published on 7 April by Stanford University’s Institute for Human-Centered AI, documents this ferment and offers a detailed picture of a technological field that is no longer the future, but a present reality demanding urgent political attention. For estrategIA, this report is essential reading**: [last year’s edition was our “recommendation of the week” in issue 30 of estrategIA](https://estrategiabyaleph.substack.com/p/estrategia-30-ia-en-el-campo-de-batalla). **In this issue, we unpack the key findings of the 2025 edition, focusing chiefly on what its more than 450 pages tell us about the intersection of AI and public power**. Chapter 6 is devoted to “Policy and Governance”, but many questions relevant to government and politics also appear elsewhere in the report.

## The regulatory explosion: from debate to concrete action

If 2023 was the year generative AI entered global public consciousness, 2024 was the year governments and legislators began responding. The AI Index reveals a dramatic increase in regulatory activity. Worldwide, mentions of “artificial intelligence” in legislative proceedings across 75 countries **rose by 21.3% in 2024 alone**, reaching nine times their 2016 level.

[![Line chart of AI mentions in legislative proceedings across 75 selected geographic areas from 2016 to 2024, ending at 1,889.](https://elcontemplador.github.io/estrategia-english/assets/images/b5708039-5062-4d1d-9741-947ede06fde9_1065x601.png)](https://substackcdn.com/image/fetch/$s_!DXpH!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fb5708039-5062-4d1d-9741-947ede06fde9_1065x601.png)

*Figure 6.2.14, original English title: “Number of mentions of AI in legislative proceedings in 75 select geographic areas, 2016–24”. Source: AI Index, 2025; chart: 2025 AI Index report. The horizontal axis runs annually from 2016 to 2024; the vertical axis is the number of mentions. The line rises from roughly 200 in 2016 to around 860 in 2018, falls slightly and levels off in 2019–20, rises in 2021, dips in 2022, then climbs sharply in 2023–24. Its explicitly labelled final value is 1,889. Intermediate values are unlabelled in the image; the approximate descriptions here are not an exact data transcription.*

In the United States, the trend is even more pronounced. At federal level, **59 AI-related regulations were introduced in 2024**, more than double the 25 in 2023, with twice as many government agencies involved. Although the passage of comprehensive federal laws remains slow — only four passed in 2024 out of 221 proposals — the regulatory machinery of executive orders and agency guidance is in full operation. Yet the true legislative epicentre in the US lies in the states: the number of state AI laws **soared from 49 in 2023 to 131 in 2024**. California leads the charge, but many other states are actively legislating, particularly on immediate concerns such as ***deepfakes***. By the end of 2024, **24 states already had specific laws** regulating or penalising the malicious use of AI-generated synthetic content, up from just five the previous year: a direct response to fears about electoral integrity and personal protection.

Internationally, the European Union reached a milestone with the final approval of its ambitious **Artificial Intelligence Act**, the world’s first horizontal, risk-based regulatory framework. Although it will take full effect progressively through 2026, its approach sets a global precedent by classifying AI systems by risk level and imposing strict obligations on those deemed “high risk”.

## Massive public investment in AI

Alongside regulation, 2024 saw an unprecedented financial commitment by states to strengthening their AI ecosystems. AI has become established as a strategic national asset, and governments are investing billions to avoid falling behind. Canada announced a **$2.4 billion** package to secure its “AI advantage”. France committed **€109 billion** to a wider technology plan prioritising AI. India launched its “IndiaAI Mission” with **$1.25 billion**. Saudi Arabia unveiled “Project Transcendence”, an ambitious **$100 billion** strategic project to establish itself as a global technology hub, including a strong commitment to Arabic-language AI models in collaboration with Google. China, meanwhile, created a new **$47.5 billion** state fund focused on advanced semiconductors, which are crucial to AI.

These astronomical sums contrast with public investment in previous years and reflect a new doctrine: technological sovereignty in AI is a matter of national security and economic competitiveness. Interestingly, the AI Index highlights differences in how that money is spent. In the United States, analysis of public contracts shows a strong concentration in the **Department of Defense —75% of spending in 2023**. In Europe, public AI investment is more evenly distributed among general public services, education and health, with defence still accounting for less than 1%.

## The geopolitics of AI: intense competition and new cooperation

AI is inevitably a new arena of geopolitics. **The United States retains its lead at the frontier of innovation**, producing 40 of the 58 notable AI models in 2024, **and attracts the most private investment**, at $109 billion. **China, however, leads in the volume of scientific output**, accounting for 23.2% of global publications, **and patents**, at almost 70% of the global total. Crucially, **it has dramatically narrowed the technical performance gap** between its models and their rivals. **By the end of 2024, the best Chinese models had reached near-parity with Western ones on key benchmarks such as MMLU and HumanEval**. Competition is fierce, as reflected in measures such as US controls on exports of advanced semiconductors to China.

[![Performance of leading US and Chinese models on LMSYS Chatbot Arena, January 2024 to February 2025; final labelled scores 1,385 and 1,362.](https://elcontemplador.github.io/estrategia-english/assets/images/41793e74-66bd-4188-968a-087d98040c95_1262x688.png)](https://substackcdn.com/image/fetch/$s_!Vbga!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F41793e74-66bd-4188-968a-087d98040c95_1262x688.png)

*Original English chart: “Performance of top United States vs. Chinese models on LMSYS Chatbot Arena”. Source: LMSYS, 2025; chart: 2025 AI Index report. Monthly dates run from January 2024 to February 2025 on the horizontal axis, with score on the vertical axis. The US line stays above China’s throughout. Both rise overall: the US series has plateaux early in 2024 and in May–July, then an October dip before rising again; the Chinese series jumps in May, is nearly flat through August, and rises strongly late in the period. The printed endpoint scores are 1,385 for the United States and 1,362 for China. Other plotted points do not have exact numerical labels.*

But 2024 was not only about competition. **International cooperation** to address shared risks also intensified. The most visible milestone was the creation of the **International Network of AI Safety Institutes**, formally launched in November 2024, bringing together bodies from the United States, the United Kingdom, the EU, Canada, France, Japan, Singapore, South Korea, Australia and Kenya. Its aim is to coordinate research on the safety of advanced models, manage synthetic-content risks and develop common standards. Similar initiatives emerged from the G7, through the Hiroshima AI Process, and the OECD, seeking a shared language and shared expectations for responsible AI.

## In search of responsible AI

Much of the political and regulatory activity in 2024 revolved around the need for responsible AI, or RAI. The principles of **transparency, explainability, fairness, safety and accountability** recur in almost every national strategy and international framework. This responds both to ethical concerns and to the pragmatic need to build public trust so that AI can be adopted.

The AI Index reveals that, although corporate awareness of RAI risks has increased — privacy and cybersecurity are the main concerns — **effective implementation of mitigation measures is lagging behind**. Only a small proportion of companies systematically carry out rigorous bias audits or impact assessments. This is where public policy has a crucial role to play, encouraging or requiring better practice.

New **technical benchmarks for evaluating models’ safety and factual accuracy** emerged, and developers’ **transparency** improved, but opaque areas remain. One emerging challenge with political implications is the **shrinking of the “data commons”**: growing restrictions on scraping public web data could limit the diversity and quality of data available to train future models, potentially affecting competition and fairness.

## Public perceptions: trust or caution?

Public opinion, the central subject of the report’s eighth and final chapter, is a key gauge of AI’s political viability. The document presents a complex picture:

- **Growing but uneven optimism:** globally, 55% believe AI will bring more benefits than drawbacks, up from 52% in 2022. Yet regional differences are enormous: 83% in China versus 39% in the United States. Strikingly, optimism has grown most in previously sceptical countries, with Germany up 10% and France up 10%.
- **Usefulness versus risk:** people value AI for saving time and for entertainment, but are wary of its impact on health, the economy and job quality.
- **Support for regulation:** both the general public and local legislators in the US show strong support for AI regulation.
- **Low institutional trust:** mistrust persists towards technology companies and their ability to protect personal data — only 47% trust them globally — and ensure fairness, with belief in AI’s impartiality declining.

[![Country comparison of agreement that AI products and services have more benefits than drawbacks, with 2022–24 survey points and printed changes reproduced below.](https://elcontemplador.github.io/estrategia-english/assets/images/1b7cbd94-2a70-493f-88dd-330e4e55ad52_1045x851.png)](https://substackcdn.com/image/fetch/$s_!bPIZ!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F1b7cbd94-2a70-493f-88dd-330e4e55ad52_1045x851.png)

*Figure 8.1.2, original English title: “‘Products and services using AI have more benefits than drawbacks,’ by country (% of total),2022–24”. Source: Ipsos, 2022–24; chart: 2025 AI Index report. The horizontal axis runs from 0% to 100% of respondents who agree. Blue points denote 2022, lilac 2023 and teal 2024. Exact values are not printed beside the survey points; the table below reproduces every country and the numerical labels in the separate “Point change 2022 vs.2024” column. Those labels describe percentage-point changes, although the image appends a percent sign. A dash means no change figure is printed, not a zero change.*

| Country | Printed change, 2022 to 2024 |
| --- | ---: |
| Argentina | 2% |
| Australia | 7% |
| Belgium | 0% |
| Brazil | −1% |
| Canada | 8% |
| Chile | −3% |
| China | 5% |
| Colombia | 2% |
| France | 10% |
| Germany | 10% |
| Great Britain | 8% |
| Hungary | 2% |
| India | −9% |
| Indonesia | — |
| Ireland | — |
| Italy | 3% |
| Japan | 6% |
| Malaysia | −2% |
| Mexico | 5% |
| Netherlands | 3% |
| New Zealand | — |
| Peru | 0% |
| Poland | −4% |
| Romania | — |
| Russia | — |
| Saudi Arabia | — |
| Singapore | — |
| South Africa | 5% |
| South Korea | 4% |
| Spain | −3% |
| Sweden | 3% |
| Switzerland | — |
| Thailand | — |
| Turkey | 9% |
| United States | 4% |

**This mix of perceptions suggests that governments have a public mandate to regulate, but must balance protection with encouraging useful innovation, while clearly communicating benefits and risks to build informed trust.**

Those are the main messages the Artificial Intelligence Index Report 2025 offers on AI, politics and government, but **the report also contains a series of highly significant general conclusions**. Its own summary is available [on its website](https://hai.stanford.edu/) and in the report’s opening pages. **Below, we offer our own condensed version**, prepared, like the rest of this article, with o3 and Gemini Pro 2.5, with the **infographic** generated using Napkin.ai.

[![Ten-panel Spanish infographic summarising the AI Index 2025; all panel text translated below.](https://elcontemplador.github.io/estrategia-english/assets/images/29ea52e9-d7db-4550-9d45-e1476a6ad34b_1600x1041.png)](https://substackcdn.com/image/fetch/$s_!6oI9!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F29ea52e9-d7db-4550-9d45-e1476a6ad34b_1600x1041.png)

*Full translation of “Key findings of the Artificial Intelligence Index Report 2025”:*

- **Acceleration and integration:** AI is deeply integrated into the economy, science and daily life. The coming years promise major changes.
- **Technical performance:** AI models consistently surpass benchmarks, even the newest and most difficult ones. Improvement is accelerating.
- **Government action:** global regulatory activity is rapidly intensifying. Multiple nations are making massive public investments in AI infrastructure and development.
- **Impact on science:** AI is revolutionising drug discovery and medicine, opening new scientific horizons and earning recognition such as Nobel Prizes.
- **Public opinion:** global sentiment towards AI is positive, though it varies greatly between countries. Concern about ethics, privacy and fairness is growing.
- **Business investment and adoption:** private investment in AI is breaking records, and corporate adoption has soared.
- **Responsible AI:** AI-related incidents and concern about ethics, bias and safety are increasing.
- **Efficiency and accessibility:** AI is becoming cheaper and more efficient, lowering barriers to entry and widening access.
- **Industrial dominance:** industry develops powerful AI models. Academia remains highly important in research.
- **Persistent challenges:** reasoning, reliability and disinformation remain challenges.

## Conclusions

The [AI Index Report 2025](https://hai-production.s3.amazonaws.com/files/hai_ai_index_report_2025.pdf) confirms that 2024 was the year the intersection of AI, politics and government reached critical mass. The speed of technical advances is forcing public actors to move with unprecedented agility. Good regulation is no longer optional, but a need perceived around the world. Public investment has become a fundamental geopolitical instrument. And international cooperation, though still emerging, is recognised as indispensable for addressing cross-border challenges such as the safety of advanced models and disinformation.

Yet **the challenges are immense. How can we regulate effectively without killing innovation in a technology that evolves every few months? How can we ensure that massive public investment delivers broad social benefits rather than narrow competitive advantages? How can we make sure responsible AI goes beyond statements of principle and is embedded in how systems are actually designed and deployed? And how can we manage the gap between AI’s technical capabilities and the understanding — and trust — of the public and many decision-makers?**

**The Stanford report offers no easy answers, but it does provide crucial data to inform the debate.** It shows a world that has decisively awoken to the AI age and is beginning to construct the rules of the game, rapidly and sometimes chaotically. The next few years will be decisive in determining whether we can forge adaptive, evidence-based, human-centred global governance that allows us to navigate the most profound transformation since the Industrial Revolution. **For estrategIA readers, the report’s message is already clear: AI is no longer merely a technical or economic matter. It is also, and very significantly, a matter of politics and global governance.**

[Fernando Nieto Lobato](https://www.linkedin.com/in/fernandonietolobato/?originalSubdomain=es)

*Director of Digital Innovation at Institución Educativa ALEPH*