Homeland Security Investigations (HSI), the investigative arm of U.S. Immigration and Customs Enforcement, is prioritizing auditability and human oversight as it expands its use of AI in criminal investigations, according to a FedScoop report. 

Peter Hatch, principal assistant director of HSI’s Office of Intelligence, outlined the agency’s requirements at a November 12 event in Washington.

Because investigative material can become evidence in court, HSI needs to verify AI-generated outputs and access the source documents behind them. Investigators must also document AI use in casework and disclose it in reports. Hatch said the agency wants tools that can support analysis while keeping human investigators responsible for validating results.

HSI already has nearly a dozen AI use cases, including summarizing and categorizing tips, classifying media evidence, and redacting audio and video. Other applications under development include lead generation, translated-text analysis, digital records review and the synthesis of fragmented maritime data.

Traceability Becomes Essential to Enterprise AI Governance

The agency’s requirements highlight a broader challenge for chief data officers (CDOs): ensuring that AI-generated insights can be traced back to reliable source data. 

Organizations deploying AI in sensitive workflows need clear records of the information supplied to models, the outputs generated and the human reviews that inform decisions.

These controls also require clearly assigned responsibilities across data, technology, security and business teams. As enterprises introduce more autonomous systems, defining ownership and accountability for AI governance becomes increasingly important.