Immigration and Customs Enforcement is developing an artificial intelligence platform designed to let teams build, test and deploy additional AI tools, expanding an automation effort that already encompasses more than 40 AI use cases across the agency.
The system, called STELLA — Secure Technology Environment for Large Language Agents — will provide developers with a secure access point to large language models and containerized environments for experimenting with retrieval-augmented generation, according to records obtained by FedScoop through a Freedom of Information Act request.
ICE said the platform will include standardized tools, authentication and authorization controls, code scanning, AI guardrails, logging and analysis, with access to large language models through the agency's FedRAMP cloud environment.
The precise deployment status remains unclear. ICE told FedScoop that STELLA was implemented June 8, but DHS continued to classify the system as pre-deployment in its July AI inventory. ICE said that once the platform is deployed, existing entries could be updated and "additional use cases will be added as development teams begin to use the platform for AI development."
ICE did not provide FedScoop with expected accuracy or hallucination rates, identify third-party AI companies involved in development, or detail what agency data could be used to train or fine-tune the system. ICE said its architecture and models were still being finalized.
STELLA forms part of a broader effort to automate ICE operations. In January, Chief Information Officer Dustin Goetz told FedScoop that the agency was "trying to automate as many of our business functions as possible," identifying lower-level cybersecurity, service-desk and administrative work as targets. "We just have to train models to help us," he said, while maintaining that human intervention would remain part of the process.
ICE's more than 40 AI use cases also include mobile-device forensics and open-source intelligence tools for identifying leads and processing tips, though the agency has retired several other AI applications.
The expansion comes alongside investments in other emerging technologies. ICE disclosed plans in August to spend between $1 million and $2 million on Boston Dynamics Spot quadruped robots for situational awareness and hazard assessment.
Privacy advocates have sought details about how data collected by the robots would be stored, how long it would be retained and whether facial recognition could be incorporated. ICE has said the robots are intended to improve safety and will not replace officers or their judgment.
