AI Act and AI Pact in Brief – EU and Artificial Intelligence

The EU and the challenge of regulating artificial intelligence without limiting its potential. Here is a summary of what you need to know about the AI Act and AI Pact. Additionally, who has signed and who has not signed the AI Pact.

European Commission

The AI Act and the AI Pact are two distinct initiatives developed within the European Union to address the challenges and opportunities related to the use of artificial intelligence (AI).

Although both deal with the regulation and management of AI, they differ in terms of objectives, implementation methods, and scope of application.

1. AI Act (Artificial Intelligence Act)

The AI Act is a legislative proposal put forward by the European Commission in April 2021. It is the first global attempt to create a comprehensive regulatory framework to govern the development, implementation, and use of artificial intelligence systems. The proposal is based on the concept of risk management and relies on a classification of AI systems into different risk categories, each requiring specific measures and controls. The primary objective of the AI Act is to protect the fundamental rights of European Union citizens, ensure the safety and transparency of AI systems, and promote responsible innovation.

Structure of the AI Act

The AI Act is based on a categorization of AI systems according to the risk they may pose to society, public safety, and citizens’ rights. The identified categories are as follows:

  • Unacceptable Risk: This category includes artificial intelligence systems that are considered dangerous and pose a concrete threat to fundamental rights or personal safety. Examples include mass surveillance systems, social scoring algorithms (similar to those used in some contexts for behavioral assessment of individuals), and other technologies that could violate democratic principles. Systems classified as “high risk” are prohibited and cannot be used within the EU.
  • High Risk: AI systems in this category are those that affect critical sectors, such as recruitment and personnel selection, education (e.g., algorithms to assess student skills), healthcare, public service management, and justice. These systems must meet stringent requirements for transparency, accuracy, and robustness. A detailed impact assessment and a compliance certification are required before their commercialization and use.
  • Limited Risk: Limited risk AI systems include technologies like chatbots, which interact with users but without serious implications. Manufacturers are required to make it explicit to users that they are interacting with an automated system and not a human, but there are no significant restrictions on use.
  • Minimal Risk: Includes low-risk AI systems, such as spam filters or content recommendation algorithms. In this case, no particular compliance measures are required.

Objectives and Impacts of the AI Act

The AI Act aims to create a balance between innovation and citizen protection. It establishes regulations that impose transparency, control, and continuous monitoring of AI systems, while creating a favorable environment for innovation, enabling European businesses to compete globally. With this legislation, the European Union aims to become a leader in “trustworthy” and “responsible” artificial intelligence.

2. AI Pact

The AI Pact is a complementary initiative promoted by the European Commission in 2023 and is configured as a voluntary agreement for companies and developers of artificial intelligence. It is a self-regulation pact, designed to fill the regulatory gap until the approval and entry into force of the AI Act. The pact invites companies to voluntarily commit to adhering to a set of ethical principles and safety standards in the use and development of AI.

Purpose of the AI Pact

The AI Pact focuses on the voluntary adoption of measures that ensure the responsible use of artificial intelligence. The initiative includes guidelines on:

  • Safety and Reliability: Ensuring that AI systems are designed and implemented to prevent physical or moral harm to users.
  • Transparency: Clearly informing users when they are interacting with an AI system and explaining how automated decisions work.
  • Non-Discrimination: Preventing the use of data that could lead to discrimination against specific groups of people or social categories.
  • Data Protection: Respecting privacy and data protection, ensuring that AI handles personal information in compliance with the GDPR (General Data Protection Regulation).

Role of the AI Pact in the European Context

The AI Pact represents an intermediate phase of self-regulation and promotion of best practices. Although it does not have legally binding value like the AI Act, it is designed to encourage companies to already implement safety and ethical measures in the use of AI, in line with the future requirements of the AI Act. The AI Pact also serves as a tool for dialogue between the private sector and European authorities, helping to develop practical solutions that could influence the final version of the AI Act.

Differences between AI Act and AI Pact

  • AI Act: It is a binding legislative proposal that, once approved, will impose specific legal obligations on companies and public entities that develop and use artificial intelligence systems. It aims to establish a uniform regulatory framework across the European Union.
  • AI Pact: It is a voluntary and non-binding initiative, designed to promote responsible and safe use of AI. It functions as a temporary and flexible agreement pending the definitive implementation of the AI Act.

In summary, the AI Act and the AI Pact represent two complementary approaches by the European Union to govern artificial intelligence, ensuring safety, transparency, and innovation while respecting European values and fundamental rights.

The European Union’s AI Pact has been signed by over 100 companies, including some of the major technology multinationals such as Microsoft, Google, OpenAI, Samsung, Vodafone, Salesforce, Mastercard, etc., as well as by small and medium-sized European enterprises. These companies represent various sectors, including computing, telecommunications, healthcare, banking, automotive, and aerospace.

The pact, which is voluntary and not legally binding, urges signatories to adopt a governance strategy to encourage the adoption of artificial intelligence within their organizations and to work towards future compliance with the AI Act. Among the companies that have not joined, Meta (formerly Facebook) is a significant absence, although the company has stated it is open to future discussions.

At present, not all major technology companies have signed the pact, but the European Commission expects that more companies may join as the full implementation of the AI Act in 2026 approaches.

Read the European Commission’s statement here >