A/75/590 country where he feared for his life, 162 in violation of non-refoulement prohibitions under international law. 51. Moreover, social media screening has compounded the disproportionate risk affecting people belonging to or presumed to be of the Muslim faith or Arab descent “by creating an infrastructure rife with mistaken inference and guilt by association”. 163 For example, in 2019, United States Customs and Border Protection, another constituent agency of the Department of H omeland Security, denied a Palestinian college student entry to the country based on his friends’ Facebook posts expressing political views against the United States, even though he did not post such views of his own. 164 In addition to the direct burdens they place on non-citizens, the expanded social media disclosure requirements of the Government of the United States foreseeably affect freedoms of speech and association. 52. Homeland Security Investigations, the investigative arm of United States Immigration and Customs Enforcement, had already been testing automated social media profiling as early as 2016, 165 strengthening its open source social media exploitation capabilities for the purposes of scrutinizing visa applicants and visa holders before and after they arrived in the United States. 166 Submissions also raised concerns about consideration by the Government of the United States of technologies whose goal was “determinations via automation” regarding whether an individual applying for or holding a United States visa was likely to become a “positively contributing member of society” or intended “to commit criminal or terrorist attacks”. 167 One submission noted in particular the use in the United States of risk assessments tools in immigration detention decisions, including one using an algorithm that was set to always recommend immigration detention, regardless of an individual’s criminal history. 168 This example is one in which technology has been tailored to pursue punitive immigration enforcement measures rooted in the racist, xenophobic and ethnonationalist vision of immigration that has been advanced by the Administration of President Donald Trump. 53. All this points to a trend in immigration surveillance where predictive models use artificial intelligence to forecast whether people with no ties to criminal activity will nonetheless commit crimes in the future. Yet, these predictive models are prone to creating and reproducing racially discriminatory feedback loops. 169 Furthermore, racial bias is already present in the datasets on which these models rely. 170 When discriminatory datasets are treated as neutral inputs, they lead to inaccurate models of criminality which then “perpetuate racial inequality and contribute to the targeting and overpolicing of non-citizens”. 171 __________________ 162 163 164 165 166 167 168 169 170 171 22/25 Ibid. Ibid. Ibid. Submission by Mijente, citing Sarah Lamdan, “When Westlaw fuels ICE surveillance: legal ethics in the era of big data policing”, New York University Review of Law and Social Change , vol. 43 (2019). Submission by Mijente, citing www.nytimes.com/2019/10/02/magazine/ice-surveillancedeportation.html. Ibid. Submission by Minority Rights Group International. Submission by Mijente. Ibid. Ibid. 20-14872