AI Translation Errors in Border Control Led to Wrongful Detentions

U.S. Customs and Border Protection (CBP) and other immigration agencies have increasingly relied on AI-powered translation tools to process the growing volume of multilingual communications and documents at ports of entry. These systems are embedded in various border control applications, including the CBP One mobile app used for asylum appointments and general translation needs during inspections. However, reports emerged in 2023 documenting systematic errors in these AI translation systems, particularly when processing Arabic, Farsi, and other Middle Eastern languages. The errors typically manifested as contextual mistranslations that created false associations with criminal activity or terrorism. For example, common Arabic phrases or names were incorrectly translated to suggest connections to illegal activities, leading border agents to flag travelers for additional screening, detention, and interrogation. In several documented cases, travelers were held for hours or days based solely on these mistranslations, with some ultimately denied entry despite having valid documentation and no actual security concerns. The problem was compounded by the lack of human oversight in the translation process. Border agents, often lacking language expertise, relied heavily on the AI-generated translations without verification. The high-pressure environment at border crossings discouraged questioning the AI outputs, and supervisory review of AI-assisted decisions was not systematically required. This created a feedback loop where incorrect AI translations directly led to enforcement actions without adequate human validation. Civil rights organizations and immigration lawyers began documenting these cases, revealing a pattern of discriminatory impact on Arabic-speaking travelers and others from Middle Eastern countries. The American Civil Liberties Union and other advocacy groups filed complaints highlighting due process violations and the disproportionate impact on specific ethnic and linguistic communities. Legal challenges focused on the lack of transparency in AI decision-making and the absence of meaningful human review before taking detention actions based on AI translations. The incidents have broader implications for AI use in law enforcement and immigration contexts. They highlight the particular risks when AI systems trained on limited or biased datasets are deployed in high-stakes government applications without adequate oversight. The translation errors appear to stem from both technical limitations in natural language processing for certain languages and potential bias in training data that associated certain terms or phrases with security threats without proper context understanding.