A/75/590 and recognition by the States in which they resided. Stateless persons, who are predominantly from racial and ethnic minorities, are systematically excluded from digital identity databases and documentation. Centralized biometric identification systems challenge the internationally recognized framework of nationality and citizenship in multiple ways. Key problems include algorithmic decision -making, and taking decisions on legal status out of the hands of government officials and placing them in the hands of machines or registrars administering biometric data kits. This can have the effect of de facto denaturalization without due process or safeguards. The same key considerations that must flow into every nationality deprivation decision, including non-discrimination, avoidance of statelessness, prohibition of arbitrariness, proportionality, necessity and legality, 58 must also be present when considering the introduction of centralized biometric identification systems. The introduction of digital governance structures risks deprivation of nationality by proxy measures, without due process – both intentionally and as a result of incomplete or flawed civil registration systems. 59 During consultations, participants from Kenyan Nubian and Somali communities, and Rohingya communities, for example, reported systematic difficulties securing digital identification, which then threatened their ability to access formal employment and other basic needs. In some cases, digital identification regimes seemed to exacerbate statelessness by resulting in complete exclusion and non-recognition of ethnic minority groups. 4. Language recognition 25. Although automated registration systems may be adopted for the purpose of enhancing bureaucratic efficiency, their technology can produce discriminatory outcomes. According to one submission, the Federal Office for Migration and Refugees, of Germany, 60 uses TraLitA, an automatic transliteration programme, to register Arabic names in the Latin alphabet. However, the system is more error -prone for applicants whose names originate from the Maghreb region, giving a success rate of 35 per cent in contrast to 85 to 90 per cent for names of Iraqi or Syrian applicants. Arabic-speaking applicants may also be subject to a dialect analysis upon registration. The Federal Office for Migration and Refugees uses software to analyse the applicant’s spoken language sample to determine the plausibility of the stated national origin. This software relies on the Levantine dialect of Arabic, 61 and the submission raises the serious concern that the software’s “susceptibility to errors has never been checked by a specialist supervisory control and cannot be understood by external actors with no recourse to the algorithms used”. 62 The obvious risk is that speakers of Arabic dialects not represented by the software may erroneously be deemed non-credible, and therefore excluded from legal and other protections on a discriminatory basis. 5. Mobile data extraction and social media intelligence on migrant and refugee populations 26. Governments are increasingly targeting the electronic devices of migrants and refugees as a means to verify the information they provide to border and immigration authorities. Officials are able to do so using mobile extraction tools that download data from smartphones – including contacts, call data, text messages, stored files, __________________ 58 59 60 61 62 12/25 Institute on Statelessness and Inclusion et al., Principles on Deprivation of Nationality as a National Security Measure (2020), available at https://files.institutesi.org/PRINCIPLES.pdf. Ibid., principle 10. Submission by Geselleschaft für Freiheitsrechte. Ibid. Ibid. 20-14872