# Marathon EU negotiations culminate in agreement on the world's first cross-sector AI law

Author: Fernando Nieto Lobato
Original publication: 2023-12-13
Spanish original: https://estrategiabyaleph.substack.com/p/estrategia-11-la-union-europea-se
English URL: https://elcontemplador.github.io/estrategia-english/essays/011/
Status: Published translation

This is a translation of the original Spanish essay published on 13 December 2023. Its claims, examples and forecasts retain that historical context.

English publication: 2026-09-29

*Historical note: this article describes the provisional political agreement as it stood in December 2023. It is not a guide to the law currently in force.*

After many months of difficult negotiations — a process that began back in 2018 and produced a concrete proposal in April 2021 — and marathon sessions over the past week, **the European Parliament and the Council of the European Union have reached a provisional agreement on the Artificial Intelligence Act, or AI Act**, **laying the foundations for what will very probably be the world's first cross-sector legislation of this kind**. All of this has taken place under Spain's rotating presidency of the EU, with considerable effort from the team led by the Secretary of State for Digitalisation and Artificial Intelligence, Carme Artigas. ([We highly recommend this interview in which she analyses the new law](https://www.elespanol.com/invertia/disruptores-innovadores/politica-digital/europa/20231212/artigas-medido-bien-prohibicion-ia-segmente-raza-creencia-orientacion-sexual/816418648_0.html).) The aim was to secure at least this major provisional political agreement before the EU presidency changed hands, leaving enough time to complete the adoption of AI legislation before the next European Parliament elections.

**The EU's Internal Market Commissioner, Thierry Breton, announced the agreement jubilantly** on 8 December, highlighting the fact that the European Union had been the first to regulate this technology.

[![Thierry Breton announces an AI Act deal with a pie chart identifying the EU as the only place with AI regulation.](https://elcontemplador.github.io/estrategia-english/assets/images/7c3aa39c-4031-4ca0-9464-132565c44601_880x1052.png)](https://substackcdn.com/image/fetch/$s_!xeh9!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7c3aa39c-4031-4ca0-9464-132565c44601_880x1052.png)

*Text visible in the original English screenshot: Thierry Breton writes “Deal!” and “#AIAct”. His graphic is headed “Continents that have an AI Regulation”; its legend labels blue “The EU” and orange “Others”, while the pie chart is entirely blue.*

With the full text of the provisional agreement still unavailable — it appears that it will not be released until January — probably the most complete account is the [press release published by the Council of the EU](https://www.consilium.europa.eu/es/press/press-releases/2023/12/09/artificial-intelligence-act-council-and-parliament-strike-a-deal-on-the-first-worldwide-rules-for-ai/) on 9 December.

We offer a brief **summary** of it and its main points, but recommend reading it in full:

The EU AI Act is a flagship legislative initiative with the potential to foster the development and adoption of safe, trustworthy AI across the EU single market by both private and public actors. **The central idea is to regulate AI according to its capacity to harm society, following a “risk-based” approach**: the greater the risk, the stricter the rules. **As the first legislative proposal of its kind in the world, it could set a global standard for AI regulation in other jurisdictions**.

This agreement, which differs substantially from the Commission's initial proposal, **focuses on high-impact general-purpose AI models and high-risk AI systems**, introducing a revised governance system with enforcement powers at EU level and extending the list of prohibitions. AI systems are classified as “high-risk”, and specific practices are prohibited, such as cognitive behavioural manipulation and the indiscriminate use of facial images. High-risk AI systems will be permitted, but subject to a set of technically feasible, less burdensome requirements.

A new governance structure is established through **the creation of an AI Office** within the Commission, advised by a scientific panel of independent experts.

**Failure to comply with the rules can lead to fines ranging from €35 million or 7% of global turnover to €7.5 million or 1.5% of turnover**, depending on the infringement and the size of the company.

To support innovation, **the provisions concerning innovation support measures have been substantially amended compared with the Commission's initial proposal**. It has been clarified that AI regulatory sandboxes must allow testing in real-world conditions, and new provisions have been added to facilitate the testing of AI systems under specific conditions and safeguards. To ease the administrative burden on smaller businesses, support measures and limited, clearly specified derogations are included.

**The provisional agreement provides that the AI Act will apply two years after its entry into force, with exceptions for specific provisions**.

## The complex and sensitive question of biometric surveillance

One of the areas that caused the greatest difficulty in the final negotiations was the use of AI for biometric surveillance:

**Among the prohibited applications, legislators agreed to ban biometric categorisation systems that use sensitive characteristics such as political, religious or philosophical beliefs, sexual orientation or race. Also prohibited are the untargeted use of facial images from the internet or CCTV cameras to create facial recognition databases; emotion recognition in workplaces and educational institutions; “social scoring” based on social behaviour or personal characteristics; AI systems that manipulate human behaviour to circumvent free will; and AI used to exploit people's vulnerabilities.**

**For law enforcement use of biometric identification systems in publicly accessible spaces, negotiators agreed on a series of safeguards and limited exceptions**, subject to prior judicial authorisation and strictly defined lists of offences. Retrospective recognition will be used strictly in targeted searches for a person convicted or suspected of committing a serious crime.

The agreed text now moves into a phase of technical refinement over the coming weeks and must subsequently be approved by Parliament and the Council to become law, with general entry into force expected in 2026.

## Views on the new law

**It is still very early, not least because the full text of the provisional agreement is not yet available. Even so, many voices have already commented on the future AI law**. Summarising the many conflicting positions we have read over the past few days is very difficult, but we want **to offer our own brief overview of the most important ones**.

[This article by Marcos Merino in Genbeta](https://www.genbeta.com/a-fondo/videovigilancia-a-competidores-chatgpt-asi-nos-afectara-nueva-regulacion-europea-inteligencia-artificial) gives a very concrete summary of some of the law's perceived main advantages and disadvantages. **Advantages** include **establishing an ethical standard for AI, through which Europe could also seek to lead this approach globally**, **as well as its distinctive risk-based approach**. Conversely, it identifies a possible **disadvantage**: the law could create **international divergences with regions such as the United States and China, which take different approaches to AI regulation, encouraging innovations to remain confined to the part of the world where they are developed and leaving the EU isolated from progress in a field that will be absolutely strategic**. It also highlights **the possibility that the obstacles created by legislating so early in a field still undergoing rapid development could hold back SMEs and entrepreneurs**, unable to bear the high costs of meeting legal requirements, **and benefit technology oligopolies and the largest companies**.

**On X, formerly Twitter, some prominent Spanish entrepreneurs**, such as Carlos Blanco and Marc Vidal, **expressed their initial opposition to the regulation**. Probably their objection was more to regulating an emerging field at all, and so early, while other countries will opt more for self-regulation and looser approaches whose overriding priority is to avoid holding back innovation and the capture of economic value.

[![Spanish posts by Carlos Blanco, Marc Vidal and WWWhatsnew criticise AI regulation and restrictions on training data.](https://elcontemplador.github.io/estrategia-english/assets/images/dcc9e0b8-e8cc-45b5-afba-1175cd45b8d6_888x751.png)](https://substackcdn.com/image/fetch/$s_!bEI6!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fdcc9e0b8-e8cc-45b5-afba-1175cd45b8d6_888x751.png)

**Text visible in the screenshot, translated from Spanish:**

> **Carlos Blanco, 9 December:** Top Spanish entrepreneurs in the USA tell me that Europe will fall behind in AI because of regulation.
>
> **Quoted post by Marc Vidal, 9 December:** Serious question… Does no one think it is wrong to regulate AI? Does no one find the overwhelming unanimity in the media surprising… [The post is truncated in this screenshot.]
>
> **WWWhatsnew, 11 December 2023:** At a recent event on AI in medicine, people were discussing how they have to buy databases from Asia to train models, because using data from here is impossible, producing results that often cannot be applied.

**The well-known Spanish entrepreneur Benjamí Villoslada, co-founder of Meneame, added a brief but significant reflection, using the comparison with a knife that many experts have raised throughout the legislative process: whether the technology itself needs to be regulated, rather than only its applications or outcomes**. That is how most technologies are treated, but not all — consider clinical trials in medicine and pharmaceuticals, or safety tests for cars. This is a very complex question, and one likely to receive different legislative approaches in different countries, in a field such as AI whose potential capabilities make it highly unusual, with a capacity for both “good” and “evil” on a scale that some experts even compare with atomic energy.

[![Marc Vidal questions AI regulation and Benjamí Villoslada replies with an analogy about knives.](https://elcontemplador.github.io/estrategia-english/assets/images/8e5410d8-29c7-4a7f-a580-24495d2fcfd3_897x550.png)](https://substackcdn.com/image/fetch/$s_!OY_s!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F8e5410d8-29c7-4a7f-a580-24495d2fcfd3_897x550.png)

**Text visible in the screenshot, translated from Spanish:**

> **Marc Vidal, 9 December:** Serious question… Does no one think it is wrong to regulate AI? Does no one find the overwhelming unanimity in the media in favour of what was approved yesterday surprising?
>
> **Benjamí Villoslada, 9 December 2023:** I think it is wrong. All the harm AI might cause is already regulated. We never introduced regulations for knives; “you shall not kill or harm anyone” is enough.

**One of the world's most authoritative voices on AI, Meta's Chief AI Scientist Yann LeCun, also commented on the legislation on X**. Like others, **he seemed particularly pleased with the changes made during the final round of negotiations in favour of open-source models**, very probably one of Europe's main routes to competitiveness in AI against the large technology companies, the vast majority of which are American. **Overall**, judging by coverage in various media — such as [this article in Expansión](https://www.expansion.com/juridico/actualidad-tendencias/2023/12/11/65773f45e5fdeae5508b4606.html?emk), which gives a voice to numerous experts — and on X and Reddit, **the new version of the law, which moderated some of the initial maximalist positions, has been considerably better received by European entrepreneurs in the sector and by researchers than previous drafts**.

[![Yann LeCun welcomes concessions for open-source AI models and quotes a report about the agreement.](https://elcontemplador.github.io/estrategia-english/assets/images/3c1ac786-9f53-4163-a0d2-1b577a0fe507_892x728.png)](https://substackcdn.com/image/fetch/$s_!_ibG!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F3c1ac786-9f53-4163-a0d2-1b577a0fe507_892x728.png)

**Transcription of the original English text visible in Yann LeCun's post:**

> The EU AI Act negotiations ended.
> One contentious issue was the regulation of foundation models, particularly open source ones.
>
> Kudos to the French, German, and Italian governments for not giving up on open source models.
>
> Juicy part:
> "The legislation ultimately included restrictions for foundation models but gave broad exemptions to “open-source models,” which are developed using code that’s freely available for developers to alter for their own products and tools. The move could benefit open-source AI companies in Europe that lobbied against the law, including France’s Mistral and Germany’s Aleph Alpha, as well as Meta, which released the open-source model LLaMA."

*The screenshot also shows a truncated Washington Post link; the complete address is not visible.*

Perhaps **the strongest criticism of the agreement came**, to many people's surprise, **from human rights groups, who were deeply disappointed by the introduction of exceptions concerning facial and biometric recognition**. They believe these exceptions could allow many practices dangerous to citizens and civil rights to slip through, and are particularly concerned that other countries will adopt these technologies on the grounds that they are not categorically prohibited even in the EU.

**Amnesty International's press release headline is revealing:**

[EU: The bloc's decision not to ban public mass surveillance in the AI Act sets a devastating global precedent](https://www.amnesty.org/en/latest/news/2023/12/eu-blocs-decision-to-not-ban-public-mass-surveillance-in-ai-act-sets-a-devastating-global-precedent/)

*Translation note: this headline is translated from the Spanish article; its precise original English wording has not been independently checked.*

**Beyond the terms of the law itself, the most common criticism on social media has concerned the EU's relevance in the AI sector and whether, beyond regulation, it will be able to lead through its vision given its near-irrelevance in business**, where efforts at development do not appear to match those devoted to legislation. Of the world's 100 largest technology companies, 60% are American, and many of them — often the largest — lead AI development, whereas Europe accounts for less than 10% of the world's 100 largest technology companies.

Many people on X therefore responded ironically to the Internal Market Commissioner's tweet announcing the agreement, offering comparisons highlighting the EU's insignificance in the technology business. It has hurried to become the first to have legislation, yet is far from having a strong AI ecosystem ([despite important advances such as those announced this past week by the French start-up Mistral, which put it at the forefront of open-source AI worldwide](https://mistral.ai/news/mixtral-of-experts/)):

[![Matteo Zullo posts a satirical two-circle chart separating AI development from the EU.](https://elcontemplador.github.io/estrategia-english/assets/images/a00c1b19-40fa-4152-87bf-83313ab89fc3_868x791.png)](https://substackcdn.com/image/fetch/$s_!uofT!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Fa00c1b19-40fa-4152-87bf-83313ab89fc3_868x791.png)

*Text and visual meaning of Matteo Zullo's original English graphic: the heading reads “Continents that have AI development”. The legend assigns orange to “Continents that have AI development” and blue to “The EU”. Two separate circles contrast the two groups, as a satirical response to Breton's graphic.*

This difference in perception is also visible in the contrast between Europe, where the news has received prominent media coverage — though perhaps not even to the extent one might expect — and elsewhere, particularly in technical AI circles. In one of Reddit's main AI communities, someone opened a thread entitled “[The AI act passed, I don't see much talk here](https://www.reddit.com/r/OpenAI/comments/18f2y9k/the_ai_act_passed_i_dont_see_much_talk_here/).”

**Perhaps the best summary for understanding this “criticism”, which goes far beyond the legislative work itself** — whose importance, at least for the EU, is generally the subject of broad agreement — is **the headline The Economist used to report the news: “Europe, a laggard in AI, seizes the lead in its regulation.”**

[![The Economist headline: Europe, a laggard in AI, seizes the lead in its regulation.](https://elcontemplador.github.io/estrategia-english/assets/images/aa2514fa-a629-498f-be24-01a3d8f2469f_1496x500.png)](https://substackcdn.com/image/fetch/$s_!hdsV!,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2Faa2514fa-a629-498f-be24-01a3d8f2469f_1496x500.png)

*Original English headline checked against the screenshot above. Its subheading reads: “The world’s first AI regulation is a bit of a mixed bag”.*

P.S. After the text of this newsletter had been finalised, we saw that [the Financial Times had reported what are probably the most critical comments on the agreement from within European politics](https://www.ft.com/content/9339d104-7b0c-42b8-9316-72226dd4e4c0), echoing the issues discussed in the preceding paragraphs. **French President Emmanuel Macron has argued that the European law could significantly hold back the sector in the EU, and in particular create a major disadvantage in a field where the United States and China are already well ahead**. The French president went so far as to say:

> “We can decide to regulate much faster and much more forcefully than our main competitors. But we will be regulating things that we will no longer produce or invent. This is never a good idea.”

*Translation note: Macron's quotation is translated from the Spanish article; its original wording has not been independently verified.*

Fernando Nieto Lobato

*Director of Digital Innovation, Institución Educativa ALEPH*