AI Ethics and the Rule of Law

An impatient future does not need perfection. Progress always needs some initial action, and the beginning is always today. The European Union’s General-Purpose AI Code of Practice came into effect on Saturday, 2 nd August 2025.

As it stands, the code is a “world-first” and undisputed best-effort starting point. It is not perfect. It will evolve. It establishes foundational norms for AI governance. The code is not legally binding. It’s a voluntary set of guiding principles designed to help providers of general-purpose AI (GPAI) models comply with the requirements of the EU AI Act, which is legally binding.

The EU AI Act entered into force on 1 st August 2024. However, its provisions will be phased in over time, with full applicability expected by 2 nd August, 2026. Some specific provisions, like those on prohibited AI practices and AI literacy obligations, started applying earlier, specifically on 2 nd February 2025. The bulk of rules for High-Risk AI Systems (HRAIS) will come into effect starting 2 nd August 2026, though some specific provisions may have later deadlines.

While not mandatory, adhering to the code can offer benefits like reduced administrative burden and increased legal certainty when demonstrating compliance with the AI Act. The code serves as a “practice direction” for GPAI platforms to meet their obligations under the EU AI Act, particularly before specific standards are fully developed. The EU AI Act itself is the legally binding framework regulating AI systems in the EU.

The code serves as a supplementary tool to assist companies in understanding the AI Act’s requirements, specifically for GPAI models. Firms can opt to follow the code or prove compliance with the AI Act through alternative methods. The European AI Office and national market surveillance authorities are responsible for enforcing the AI Act, not the code itself.

The code establishes clear expectations: be transparent about training data, assess risks such as bias or misinformation, and explain how the model works. OpenAI and Mistral have added their signatures to the EU AI Code.  Google announced that it intends to sign the code. Developers of AIs appreciate that over-regulation can asphyxiate innovation, arrest development, and hinder the diffusion of AI technologies.

Some developers are uncomfortable with the vagueness of the code. They argue that it exceeds what the AI Act will eventually require, and that the nebulousness creates legal uncertainties. Digital governance tools like this code are templates for countries interested in drafting and applying policies, protocols, and standards for the infosphere’s judicious development, use, and management.

As we move towards Superintelligence, governments will need softlaws to (1) determine and control the methods and processes used by data curators to improve data quality, reliability, access, security, and availability, and (2) devise protocols for decision-making and data accountability.

Many private actors advocate self-regulation in the ethics of AI and lobby against the introduction of norms. This weakens the enforcement of legal standards and has sweeping effects at all levels of society and across every sector of the economy where ethics is exploited as an alternative to the rule of law. One way or the other, legislation catches up.

Short-term lobbying builds a barricade to control the flow of the legislative agenda, making compliance bearable but misaligned with legal thought. The proliferation of principles and ethical frameworks facilitates the export of dishonourable practices.

Bad actors are more likely to engage in ethics dumping in contexts where disadvantaged populations, weaker institutions, unfair power distributions, and other social, political, and economic ills carry the day. What is needed is a global ethic of distributed responsibility.

Traditionally, discrimination is defined as the differential and harmful treatment of individuals or groups based on arbitrary categorisations such as origin, culture, disability, sexuality, and gender identity. This problem surfaces in various forms in multiple spheres, with the infosphere being one of the areas where its impact is worrying. In the digital domain, discrimination can manifest as biases in algorithms and exclusionary practices that restrict access.

These manifestations not only mimic existing inequalities in society but also have the potential to intensify them, limiting access to prospects and reinforcing negative labels. To effectively address discrimination in the digital environment requires a comprehensive approach that considers both individual and collective accountability, and engages users, developers, and regulators with practice guidelines or codes.

The novel convergence of AI and ethics brings into relief the need to address the moral consequences of developing and using digital artefacts. This necessity calls for a profound understanding of digital ethics, which includes data privacy, cybersecurity, digital inclusion, and the impact of AI energy consumption patterns on the climate emergency.

Inherited pre-judgements and prejudices from the European Enlightenment can be amplified in digital environments, which can limit access to opportunities and reinforce negative groupings.

To address these plights effectively, it is essential to adopt a comprehensive approach that engages citizens, developers, and regulators of AI-Assemblages. The new digital anthropologies underscore the value of individual dignity and the right to autonomy emphasising privacy and the ethical management of personal information, and highlighting the importance of consent and data protection.