Building A Facial Recognition Accountability Checklist For The Public Sector

Nations advancing in technology and leading digital transformations must prioritize Al deployment respecting human rights. This policy brief suggests an assessment toolkit for countries to align with G20 goals, fostering Al innovation responsibly and respecting fundamental and human rights. Influenced by international experiences, our proposal addresses facial recognition technologies (FRT), aiming to strike a balance between innovation and individual and group rights protection. In light of the evolving landscape of technological advancement and the increasing deployment of facial recognition technologies (FRT), it is imperative to establish a comprehensive and responsible framework that aligns with human rights principles. We draw inspiration from previous T20 intelligence and other sources, including Idec’s and InternetLab’s Guide for Private Sector Use of Facial Recognition, Access Now’s report “Bodily harms: how Al and biometrics curtail human rights”, IFF’s Project Panoptic, and EDRi’s report The Rise and Rise of Biometric Mass Surveillance in the EU, adapting them to the context ofpublic sector implementation. Countries’ experiences will also inform our work, such as USA’s NTIA and the Council of Europe. This synthesis enables us to present a text aimed at enhancing the governance of FRT in the public domain. Recognizing FRT risks, we advocate for responsible adoption and participative data governance. Our framework, integrating case studies and principles, offers a comprehensive guide for the G20, focusing on sustainable and inclusive digital development. While acknowledging challenges in public spaces, our criteria development for FRT in public services emphasizes, for instance, impact assessments to prevent discrimination. In summary, our submission distills international FRT experiences, deriving meaningful lessons, and proposes a forward-thinking checklist for the public sector’s standards in FRT use and procurement, prioritizing non-use where necessary.