The EU introduces the pioneering EU AI Act, aiming to become a global hub for human-centric, trustworthy AI.

Laws/Regulations directly regulating AI (the “AI Regulations”)

The primary legislative proposal for regulating AI in the EU is the EU AI Act (here) and current text (here). The EU has also proposed the AI Liability Directive (here) which is designed to ensure that liability rules are appropriately applied to AI-related claims.

Status of the AI Regulations

The EU AI Act has now been finalized and was endorsed by all 27 EU Member States on February 2, 2024, and by the European Parliament on March 13, 2024. It is now awaiting a vote by the Council of the EU, before being published in the EU’s Official Journal. The EU AI Act will enter into force on the 20th day after publication, and will be effective after 24 months,¹ except for the specific provisions listed in Article 85(3). The official entry into force is expected between late June and mid-July of this year.²

The AI Liability Directive is in draft form and is yet to be considered by the European Parliament and Council of the EU.³ Timing remains uncertain.

Related laws affecting AI

There are many laws applicable in the EU that may affect the development or use of AI in the EU. A non-exhaustive list of common examples includes:

• The EU General Data Protection Regulation (EU) 2016/679
• The Product Liability Directive, which, if adopted, will allow people harmed by software (including AI software) to receive compensation from the software manufacturer (replacing Directive 85/374/EEC)
• The General Product Safety Regulation 2023/988/EU, replacing Directive 2001/95/EC
• Various intellectual property laws under the national laws of EU Member States

Definition of “AI”

AI is defined in the EU AI Act using the following terms:

• “AI system” means “a machine-based system designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment and that, for explicit or implicit objectives, infers from the input it receives how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments”
• “General purpose AI model” means “an AI model, including when trained with a large amount of data using self-supervision at scale, that displays significant generality and is capable of competently performing a wide range of distinct tasks regardless of the way the model is placed on the market, and that can be integrated into a variety of downstream systems or applications. This does not cover AI models that are used before release on the market for research, development and prototyping activities”⁴
• “General purpose AI system” means “an AI system that is based on a general purpose AI model, that has the capability to serve a variety of purposes, for direct use as well as for integration in other AI systems”

The AI Liability Directive will likely adopt the same definition as the EU AI Act.⁵

Territorial scope

The EU AI Act applies extraterritorially to:⁶

• Any provider placing, or otherwise putting into service, an AI system or general purpose AI models on the EU market, regardless of whether the provider is established or located within the EU or in a third country
• Any deployers of AI systems who have their place of establishment in, or who are located in, the EU
• Any provider or deployer of an AI system that have their place of establishment or are otherwise located in a third country, if: (i) the output produced by the AI system is intended to be used in the EU; or (ii) where an operator performs an activity by an AI system that qualifies as high risk⁷

The AI Liability Directive applies to non-contractual fault-based civil law claims within the EU.⁸

Sectoral scope

The EU AI Act is not sector-specific. It applies to all sectors.

The AI Liability Directive is not sector-specific. It applies to non-contractual fault-based civil law claims brought before national courts.

Compliance roles

Under the EU AI Act:

• Any developer of an AI system or general-purpose AI model, or any natural or legal person, public authority or other entity that has an AI system or general-purpose AI model developed and places them or puts the system into service on the EU market are “providers” under the AI Act
• Any natural or legal person, public authority or other entity in the supply chain that is not a provider or importer and makes an AI system available on the EU market is a “distributor” under the AI Act⁹
• Any natural or legal person, public authority or other entity located or established in the EU that placed on the market an AI system that bears the name or trademark of a natural or legal person, public authority or other entity established outside of the EU are “importers” under the AI Act¹⁰
• Any natural or legal person, public authority or other entity that uses an AI system, except in the course of a personal non-professional activity, are “deployers” under the AI Act¹¹
• Any provider, product manufacturer, deployer, importer, distributor or authorized representative are “operators” under the AI Act¹²

Each of these roles comes with a set of compliance obligations.

The AI Liability Directive would increase the likelihood of a successful claim against an AI system developer or the user of an AI system that relied on its output.¹³

Core issues that the AI Regulations seek to address

The EU AI Act is intended to promote the uptake of human-centric and trustworthy AI and to ensure a high level of protection of health, safety, fundamental rights, democracy, and rule of law from harmful effects of AI systems while supporting innovation and the functioning of the internal market.¹⁴

The AI Liability Directive aims to ensure that persons harmed by AI systems enjoy the same level of protection as persons harmed by other technologies in the EU. Current fault-based liability rules are not suited to handling liability claims for damage caused by AI-enabled products and services. Specifically, it may be difficult (or prohibitively expensive) for victims to prove the fault of a potentially liable person, and/or the causal link between the fault and the damage suffered, owing to the complexity, autonomy and opacity of AI systems.

Risk categorization

The EU AI Act classifies AI systems, and imposes requirements, according to different levels of risk:

• Unacceptable risk: AI systems that present an “unacceptable” risk are prohibited. This includes (among others) AI systems used for social scoring and AI systems that use deceptive or exploitative techniques to materially distort a person’s behavior in a manner that can cause harm.¹⁶
• High risk: AI systems that present a “high” risk include AI systems. These fall within two categories: (i) AI systems used as a safety component of a product (or otherwise subject to EU health and safety harmonization legislation); or (ii) AI systems deployed in eight specific areas, including (among others) education, employment, access to essential public and private services, law enforcement, migration, and the administration of justice.¹⁷
• Limited risk: AI systems that present “limited” risk include those that directly interact with natural persons (e.g., chatbots), emotion recognition systems, biometric categorization systems, and AI systems that generate "deep fakes" (i.e., audio or visual content that appears genuine, even though it is created by an AI system) are required to disclose the fact that the content has been artificially generated or manipulated.¹⁸ The transparency obligations imposed on deployers of these AI systems do not apply where the use is authorized by law to detect, prevent, investigate and prosecute criminal offenses. If the content is "evidently" an artistic, creative, satirical, fictional analogous work or program, these obligations are limited to the disclosure of existence of "deep fakes" in an appropriate manner that does not hamper the display or environment of the work.¹⁹
• Low or minimal risk: Any AI system not caught by the above are of low or minimal risk.²⁰

The AI Liability Directive does not directly govern the risks posed by AI systems.

Key compliance requirements

Compliance obligations are primarily determined by the level of risk associated with the relevant AI system:

• Unacceptable risk: AI systems posing an unacceptable risk are not subject to compliance requirements; they are prohibited outright
• High risk: AI systems and their providers (or where applicable, the authorized representative) must be registered in an EU database before being placed onto the EU market or put into service, and must comply with a wide range of requirements on data training and data governance, technical documentation, recordkeeping, technical robustness, transparency, human oversight, and cybersecurity²¹
• Limited risk: Providers and deployers of certain AI systems and general-purpose AI models are subject to transparency obligations²²
• Low or minimal risk: AI systems do not have specific obligations or requirements under the EU AI Act²³

The EU AI Act also provides for the development of codes of conduct for AI systems, which the Commission hopes all AI system providers will voluntarily apply.²⁴

The AI Liability Directive does not contain compliance requirements.

Regulators

Enforcement of the EU AI Act involves a combination of authorities, and the position is not settled. EU Member States will establish or designate at least one notifying authority and at least one market surveillance authority (together, the “national competent authorities”) and ensure that the national competent authorities have adequate technical, financial and human resources, and infrastructure (that are sufficiently knowledgeable) to fulfill its tasks under the EU AI Act.²⁵

The notifying authority is responsible for setting up and carrying out the assessment and designation procedures that are required under the EU AI Act, in an objective and impartial manner.²⁶

The market surveillance authority may vary for “high” risk AI systems, AI systems used by financial institutions subject to EU legislation on financial services, and other EU institutions, agencies, and bodies.²⁷

The market surveillance authority is primarily responsible for enforcement at the national level.²⁸ If an AI system is non-compliant, the market surveillance authorities can exercise the enforcement powers described below. The market surveillance authorities will report to the Commission and relevant national competition authorities on an annual basis.²⁹

National courts of EU Member States will be responsible for implementing the AI Liability Directive in the case of non-contractual fault-based civil law claims brought before them.

Enforcement powers and penalties

Where the market surveillance authority finds that there is: (i) non-compliance with the obligations of the EU AI Act; or (ii) compliance from a high-risk AI system with the obligations of the EU AI Act, but still presents a risk to the health and safety of persons, the fundamental rights of persons, or other aspects of public interest protection; then the relevant market surveillance authority can (a) require the relevant operator to take all appropriate corrective actions (in the event of (ii), to ensure the AI system concerned no longer presents that risk) or withdraw/recall the AI system from the market; or (b) where the operator fails to do so, the relevant authority shall prohibit/restrict the AI system being made available on its national market or put into service, or withdraw/recall the product or the standalone AI system from the market.³⁰

Penalties range from (i) the higher of €35,000,000 or up to 7 percent of a company’s total worldwide annual turnover for non-compliance with prohibited AI practices, to (ii) the higher of €7,500,000 or up to 1 percent of a company’s total worldwide annual turnover for the supply of incorrect, incomplete, or misleading information to notified bodies and national competent authorities.³¹

The AI Liability Directive increases the claimants’ likelihood of a successful claim by creating a rebuttable presumption of causality on the defendant. In practice, the new rule means that if a victim can show that someone was at fault for not complying with a certain obligation relevant to their harm, and that a causal link with the AI performance is reasonably likely, the court can presume that this non-compliance caused the damage.³²

The AI Liability Directive also gives national courts the power to order disclosure of evidence about high risk AI systems that are suspected of causing damage, to help victims access relevant evidence to identify the person(s) that could be held liable.³³

1 See EU AI Act, Article 85(1)&(2).
2 See EU AI Act, Article 83.
3 See Procedure File: 2022/0303(COD) | Legislative Observatory | European Parliament (europa.eu).
4 See EU AI Act, Articles 3(1), 3(44b) and 3(44e).
5 See AI Liability Directive, Article 2(1).
6 See EU AI Act, Articles 2(1)(a) to (c). Responsibilities along the AI value chain (including distributors, importers, deployers) are set out in Article 28.
7 See EU AI Act, Recital 11.
8 See AI Liability Directive. Article 1(2).
9 See EU AI Act, Article 3(7).
10 See EU AI Act, Article 3(6) .
11 See EU AI Act, Article 3(4).
12 See EU AI Act, Article 3(8).
13 See AI Liability Directive, Article 4(b).
14 See “Purpose” in the Procedure File: printficheglobal.pdf (europa.eu); and EU AI Act, Article 1(1)
15 See EU AI Act, Recital 88.
16 See EU AI Act, Article 5.
17 See EU AI Act, Article 6 and Annex III.
18 See EU AI Act, Articles 52(1) to 52(3)(2).
19 See EU AI Act, Article 52(3).
20 See page 4 of the briefing note.
21 See EU AI Act, Articles 8-15 and 51.
22 See EU AI Act, Article 52.
23 See page 4 of the briefing note.
24 See EU AI Act, Title IX (Codes of Conduct).
25 See EU AI Act, Article 59.
26 See EU AI Act, Article 30.
27 See EU AI Act, Article 65.
28 See EU AI Act, Article 65.
29 See EU AI Act, Article 63.
30 See EU AI Act, Articles 65(2), 65(5) and 67(1).
31 See EU AI Act, Articles 71(3) and (5) .
32 See AI Liability Directive, Article 4(1).
33 See AI Liability Directive, Article 3(1).

Jeffrey Shin (Trainee Solicitor, White & Case, London) and Daniel Mair (Trainee Solicitor, White & Case, Paris) contributed to this publication.

[View source.]