The EU’s Contested Risk-Based Approach for the Regulation of Artificial Intelligence – Trust in AI Applications and Issues of Contestation

In April 2021, the European Commission published a proposal for a Regulation on Artificial Intelligence (COM(2021) 206 final) that has been intensively debated in the European Parliament, the Council, in academia and in civil society since then. The Commission aims at supporting the development of trustworthy AI applications, opting for a risk-based approach. Academia and practitioners mostly welcomed this approach, but the details are highly contested. Policing and criminal justice are not exempted from the scope of the AI regulation, but they may use specific “backdoors” even for risky AI applications that would be prohibited or restricted for other state agencies and non-state actors. With the inclusion of security agencies, the draft AI Act is also an attempt to overcome exceptions and opt-outs that characterised the European Union’s Area of Freedom, Security and Justice (AFSJ) in policy areas such as data protection. The paper situates these “backdoors” in the context of normative requirements for the use of AI in a democratic rule of law context, namely trustworthiness, fairness, transparency and explainability.