Accountability Principles for Artificial Intelligence (AP4AI): A Universal Framework for the Internal Security Domain

Prof. Saskia Bayerl (CENTRIC)
Prof. Babak Akhgar (CENTRIC)
Dr. Gregory Mounier (Europol)

Teresa Magno (Eurojust)

Nienke van der Have (FRA)

Vanja Skoric (ECNL)

Wenche Medin (Swedish National Police)
 

AP4AI (Accountability Principles for Artificial Intelligence) is a project led by Europol and CENTRIC, in partnership with four European Union Agencies: the EU Fundamental Rights Agency, Eurojust, Cepol and the EU Asylum Agency, coordinated via the EU Innovation Hub for Internal Security.

The project has developed a comprehensive global Framework for AI Accountability for Policing, Security and Justice. The AP4AI Framework is operationally grounded and provides concrete and pragmatic support to practitioners in all stages of the AI lifecycle (design, procurement, deployment, modification, etc.). The AP4AI Framework is based on 12 Accountability Principles for AI use in the internal security domain, which have been developed and empirically verified by internal security practitioners. These AI Accountability Principles have also been validated by the consultation of more than 6, 000 citizens in 30 countries.

The AP4AI project partners have developed a series of practical considerations for the implementation of the 12 AP4AI Accountability Principles, including materiality thresholds, applicable laws, human rights and data protection impact assessment, as well as implementation guides.

AP4AI proposes to include these practical implementation measures and processes into AI Accountability Agreements (AAAs).  AAAs should be viewed as social contracts underpinned by legal obligations between internal security organisations and their stakeholders including citizens, oversight bodies and suppliers. AAAs will drive the practical implementation of the AP4AI Principles and their realisation in an operational settingand across different application areas of AI within the internal security domain.

With this, AP4AI delivers a universal mechanism that allows practitioners to assess, review, and demonstrate AI accountability systematically, reliably and pragmatically. The ambition of the project is also to contribute to the practical implementation of the future European AI Act in the internal security domain.The aim of the session is to present the AP4AI framework and discuss with participants how AI Accountability Agreements can contribute to the implementation of the future EU AI Act in the internal security domain and whether this model can become a universal accountability framework for the internal security domain.

While Eurojust will provide the views of the justice sector on the accountability framework, the police perspective will be provided by Europol.