ICE Signs $12.2 Million AI Surveillance Contract

ICE purchased Project SAFE HAVEN from Edge Ops LLC under a $12.2 million contract that promises AI-driven mapping of individuals' daily routines, real-time location tracking, and automated categorization of people and groups as potential threats. The procurement names the Homeland Security Task Force National Coordination Center as the intended user, signaling interagency operational use.

The procurement documents describe a "question-based AI interface" and persistent passive data collection to generate "patterns of life" and "target profiles." Vendor materials and the contract cite capabilities that include:

• •real-time geolocation and route reconstruction
• •linking activity across Wi-Fi networks and mobile smart devices
• •pattern-of-life analytics to identify habitual locations and routes
• •automated classification of affiliations (gangs, cartels) and threat-scoring

No public technical spec, model architecture, or evaluation metrics accompany the purchase. From a practitioner perspective, the likely components are entity resolution, graph analytics for relationships, time-series geospatial inference, and supervised classifiers that combine heterogeneous signals. Key unknowns are the data sources, labeling processes, bias-mitigation, and red-teaming or adversarial testing regimes.

This contract is a vivid example of how federal procurement funds are accelerating the operationalization of surveillance AI across immigration enforcement. The DHS has expanded discretionary acquisition authorities in recent years, enabling large contracts that integrate sensor-derived signals, commercial geolocation metadata, and device telemetry into automated pipelines. For ML teams and engineers, this raises questions about dataset provenance, measurement error in Wi-Fi and mobile-derived location signals, feedback loops from enforcement-driven labels, and the lack of public benchmarks for performance and false-positive rates in high-stakes settings.

Automated "patterns of life" systems amplify harms if classification errors map to detentions, deportations, or stigmatization. Passive collection of device signals creates broad coverage with limited consent and weak legal safeguards. Without transparent testing, thresholds, or independent audits, model drift, racial or socioeconomic bias, and correlated sensor failure modes can produce systemic harms.

Requests for oversight via FOIA, congressional inquiries, inspector general audits, and any public release of technical specifications or evaluation reports. Practitioners should follow whether the contractor discloses datasets, model types, performance metrics, and whether any external audits or red-team exercises are mandated.