Skip to main content

New regulation on the horizon – understanding the proposed EU AI Act

Artificial intelligence (AI) is becoming an instrumental part of human society, revolutionizing many aspects of life. AI adoption is now more widespread and no longer limited to high tech or specialised industries such as banking. With the many benefits AI creates, there remains a risk that it could be utilised in negative ways, some of them being manipulative and unlawful, aiming to harm and defraud people.

Inevitably, questions around risks and ethics come to light. To fill this void, address the rising concerns and harness the power of AI, several regulatory initiatives have emerged, a prominent one being the proposed EU Artificial Intelligence Act1 . It is the culmination of years of continuous discussions and consultations among regulators, industry participants, scientists and the wider public. Its scope is vast and will have a seismic impact on the many industries and entities which it will affect.

This blog introduces the key considerations within the EU AI Act and sets the scene for a series of subsequent blogs exploring the main principles of the proposal.

De-coding the act - creating a ‘trustworthy’ AI environment

Prior to the EU releasing the proposal for an AI Act, an independent high-level expert group on AI set up by the European Commission produced a study on the ethics guidelines for trustworthy AI2. The principles outlined in this report are based on three fundamental cornerstones – lawfulness, ethics, and robustness. By endorsing this framework, the group aimed to facilitate an environment of transparency, inclusion, and fairness where technology is not used to manipulate and harm people and societies.

Risk Tiers and Main Pillars

The proposed EU AI Act considers the guidelines for trustworthy AI to frame the requirements for different types of AI systems based on their risk profiles – namely, prohibited, high-risk and low-risk AI systems. Some uses of AI, such as social scoring and dark pattern AI, are prohibited unless they are used for a justifiable reason by the authorities such as solving a criminal case3.

The definition of High-risk AI is outlined in Annex III of the Act and includes systems intended as a safety component of products, those operating in areas such as biometric identification, law enforcement, employment, migration and other critical areas - such as evaluating the creditworthiness of natural persons.

Providers of such High-risk AI systems must comply with the main pillars of the proposed act:

  • Data and Data Governance – training, validation and testing datasets shall be relevant and high quality, free of error and bias4
  • Technical Documentation and Record Keeping – documentation of all technical requirements and automatic recording of events (‘logs’)5
  • Transparency – to enable users to interpret the system’s output and use it appropriately (e.g. user guides)6
  • Human Oversight – AI systems to be overseen by natural persons to prevent or minimise the risk to health, safety, or fundamental rights7
  • Accuracy, Robustness and Cyber Security - resilient in regards to errors, faults or inconsistencies within the system or the environment in which the system operates8

Conformity Assessments and Codes of Conduct

In simple terms, the proposal translates into firms being required to enhance their risk management systems and comply with each of the above-mentioned principles. Entities that are engaged in developing, deploying and using AI systems, especially high-risk ones, should have clearly defined methodologies and practices in place. Not only should firms produce detailed technical documentation, but they should also undergo conformity assessments of their internal controls to prove compliance. This will be voluntary for the providers of low-risk AI systems; however, the EU Commission will encourage and facilitate the drawing up of codes of conduct that are intended to cover the above-mentioned requirements.

Regulatory Scope

Like other European Union legislation, the principle of extraterritoriality adds another dimension of complexity – the AI Act will cover not only EU companies but also entities from abroad that provide AI services to EU citizens. Undoubtedly, this increases the prominence of the Act and serves as a reminder for companies across the globe that they must enhance their AI practices and adhere to the EU rules, notably as the Act also brings the prospect of significant fines for non-compliance.


The scene is set, and this proposal is a clear indication of the future direction of regulatory scrutiny. Authorities are applying greater pressure on companies over their AI systems and it is essential that AI providers and users have robust risk management frameworks, comprehensive controls and validation methodologies in place. Firms must re-examine and, where necessary, adapt and enhance their internal practices to the new governing practices of the EU AI Act. This will be a step towards harnessing the power of AI in a positive way, and importantly, help manage risk.

At Deloitte, we appreciate the difficulties of navigating through the proposed EU AI Act. Considering the complex global regulatory environment, we have developed a comprehensive approach to validate your algorithms and AI systems to safeguard compliance and improve internal controls. We invite all market participants who wish to discuss the AI Act, its implications, and our Algorithm and AI approach, to get in touch.



1 EUR-Lex - 52021PC0206 - EN - EUR-Lex (

2 Ethics Guidelines for AI (

3 Article 5 of the proposal

4 Article 10 of the proposal

5 Articles 11 and 12 of the proposal

6 Article 13 of the proposal

7 Article 14 of the proposal

8 Article 15 of the proposal


Proposal and Annex to the Proposal for a Regulation of the European Parliament on Artificial Intelligence: EUR-Lex - 52021PC0206 - EN - EUR-Lex (

Independent High-Level Expert Group on Artificial Intelligence – Ethics Guidelines for Trustworthy AI: Ethics Guidelines for AI (

Sign up for the latest updates