2025年4月17日，ICE紧急寻求用于“移民全生命周期”执法的AI模型，并计划在9月底前由Palantir交付ImmigrationOS原型。该推进直接服务于白宫每年100万起遣返目标，显示执法技术化正在以明确时间表和高配额加速。

在特朗普第二任期开始后的13个月内（至2026年2月14日部分停摆前），DHS已授出约12亿美元IT合同，其中超过8100万美元流向Palantir，约占6.8%。同期数据整合范围持续扩张，从车辆、电话、公用事业和社交媒体延伸到社保、车管、银行可疑活动报告与福利系统，呈现跨库联接和自动筛查的统计性上升趋势。

AI还把取证流程从每份司法文书2至3天压缩到不足1小时，意味着理论上可实现约48至72倍的处理提速，并可能显著推高令状申请量。与此同时，误报与不透明风险在扩大，叠加ICE可调取超过2000个与Ring合作的地方部门视频资源及Clearview超700亿张图像数据库，促使法院限制共享并引发更强法律反制。

On April 17, 2025, ICE urgently sought an AI model for “end-to-end immigration lifecycle” enforcement, with a Palantir ImmigrationOS prototype due by late September. The push aligns directly with the White House target of 1 million deportations per year, indicating accelerated, quota-driven technologization of enforcement.

In the 13 months after Trump’s second term began (before the February 14, 2026 partial shutdown), DHS awarded about $1.2 billion in IT contracts, with more than $81 million to Palantir, roughly 6.8% of the total. Over the same period, data fusion expanded from vehicle, phone, utilities, and social media records to Social Security, DMV, bank suspicious-activity reports, and welfare-system data, showing a rising trend in cross-database matching and automated vetting.

AI is also compressing warrant-preparation time from 2-3 days to under 1 hour per request, implying a theoretical 48-72x throughput gain and likely higher warrant volume. At the same time, false-positive and opacity risks are increasing, while ICE can draw on footage from 2,000+ local departments linked to Ring and Clearview’s 70+ billion-image database, intensifying court limits and legal pushback.

Source: How ICE’s new software tools could speed up deportations

Subtitle: The risk of overreach is high

Dateline: 2月 19, 2026 04:53 上午