Nvidia, the computing giant that this week became the world’s first $5 trillion company, is powering U.S. Immigration and Customs Enforcement’s investigative division, according to federal records reviewed by The Intercept.

This summer, ICE renewed access to software tools for use by Homeland Security Investigations, or HSI, an enforcement division previously tasked with transnational crime that has become increasingly common on American streets under the Trump administration.

The $19,000 transaction, according to federal procurement data, provided “Nvidia software licenses, which will be used by Homeland Security Investigations to enhance data analysis & improve investigative capabilities through high-performance computing solutions.”

“HSI’s growing investment in LLMs” — large language models — “suggests that it may be investing in systems that can be used to surveil U.S. citizens, migrants, and visitors,” said Amos Toh, senior counsel at the Brennan Center for Justice.

Large language models can be used to draw inference by fusing people’s publicly available data, and might be used by ICE to “to identify persons of interest and generate investigative leads.” There are well-documented flaws, however, in the way the AI crunches data and reproduces biases.

Toh said, “These problems make it more likely that people will be targeted based on flawed intelligence.”

In a statement, ICE said, “Like other law enforcement agencies, ICE employs various forms of technology to investigate criminal activity and support law enforcement efforts while respecting civil liberties and privacy interests.”

When asked whether Nvidia had any ability to ensure ICE was using its technology lawfully, company spokesperson John Rizzo told The Intercept, “Millions of U.S. consumers, businesses, and government agencies use general-purpose computers every day. We do not and cannot monitor the use of general-purpose computers by U.S. government employees.”