The recent success of Chatbots such as ChatGPT, has led to much debate on the use of artificial intelligence (“AI”) in the financial sector. Chatbots largely owe their rise in popularity to their use of machine learning, which is a subclass of AI allowing for the speedy gathering and analysis of data.
The AI Act of the European Union
With increasing performance and wide availability, the use of AI has led in 2021 to the draft of a proposal by the European Parliament and the Council for a regulation laying down harmonised rules on AI, and which has now been adopted by the European Parliament as of 13 March 2024 (the “AI Act”). This AI Act should form part of a broader AI Package, which will include additional policy measures to support the development of trustworthy AI, such as the Coordinated Plan on Artificial Intelligence which aims to accelerate investment in AI, implement AI strategies and programmes and align AI policy to prevent fragmentation within Europe.
Main Features of the AI Act
The AI Act aims at providing a long-lasting legal basis for the use of AI. The goals of the AI Act are to protect fundamental rights, harmonise the legal environment for AI at the European level and foster innovation with the European Union.
To this effect, the AI Act contains several distinct features, the most relevant of which we listed below.
1. Integration into existing legal texts
For the sake of simplicity, the AI Act should serve as the basis for legislation on AI at the European level and be integrated in the existing legal framework of each Member State.
Regarding financial and banking regulation, we can read in Recital 158 of the AI Act, that the competent authorities for the supervision and enforcement of legal acts such as regulation (EU) No 575/2013 on prudential requirements for credit institutions and investment firms should be designated, within their respective competences, as competent authorities for the purpose of supervising the implementation of the AI Act.
Overall, the approach of the AI Act is to minimise disruption, including in financial regulation, and rely on existing rules and institutions.
2. Creation of a classification system based on risk
The classification system with regard to AI systems can be summarised as shown in the following diagram:
Firstly, the AI Act prohibits a variety of uses for AI systems in its Article 5. As an example, AI systems that deploy subliminal technique beyond a person’s consciousness shall be prohibited, as they are considered to pose an unacceptable risk to the public.
Secondly, the AI Act sets-up the category of high-risk AI systems. Article 6 of the AI Act creates two criteria for the classification which are:
- The use of the high-risk AI system as a safety component or as standalone safety tool for use covered by the Union harmonisation legislations listed in Annex I of the AI Act; and
- The product which uses a high-risk AI system as a component, or the AI system in itself is required to undergo a third-party conformity assessment. The AI ACT specifies that this conformity assessment is also related to the harmonisation legislations, listed in Annex I of the AI Act.
The AI Act creates basic transparency requirements which shall apply to certain AI systems, such general-purpose AI systems. As an example, Article 50 of the AI Act states that AI providers shall ensure that certain AI systems intended to interact with natural persons are designed and developed in such a way that natural persons are informed that they are interacting with an AI system, unless this is obvious from the circumstances and the context of use.
3. Law enforcement exemption
Article 5 states that the use of “real time” remote biometric identification systems by law enforcement is normally prohibited. The AI Act does; however, introduce a few exemptions for (i) specific victims of abduction or trafficking, (ii) the prevention of specific, substantial, and imminent threat to the life or physical safety of natural persons, or (iii) the localisation or identification of a person suspected of having committed a criminal offence.
4. New Supervision structures
At the European level, the AI Act establishes a European Artificial Intelligence Board (“EAIB”).
In a similar fashion to existing European agencies such as ESMA, the EAIB will be tasked with coordinating the action of the various European competent authorities in implementing the AI Act and will provide support to the Commission and the national supervisory authorities in how to analyse and react to emerging issues in the field of AI.
5. Creation of regulatory sandboxes
As the use of AI systems is developing rapidly, the goal of the AI Act is not only to protect EU citizens but also to foster innovation in AI at the European level. To this effect, the AI Act has developed the concept of regulatory sandboxes.
Article 57 of the AI Act states that a regulatory sandbox shall provide “for a controlled environment that fosters innovation and facilitates the development, training, testing and validation of innovative AI systems for a limited time before their being placed on the market or put into service pursuant to a specific sandbox plan agreed between the prospective providers and the competent authority.”.
Going forward
The full of effect of this legislation and its practical implementation cannot be assessed yet. The creation of the EAIB for one, will undoubtedly give rise to a wide variety of guidelines and opinions related to the AI Act, thus contributing to the expansion of legal doctrine pertaining to AI technology in Europe.
In order to foster early implementation of the measures foreseen by the AI Act, the European Commission has initiated an AI Pact, which will encourage companies to share the processes and practices they are putting in place to ensure compliance to the AI Act. Companies will also be encouraged to pledge to work towards compliance with the AI Act.
Overall, the AI Act aims to be a future-proof legislative approach, which will be able to adapt to the rapidly changing environment of AI technology.
