The AI Tools CBP Is Using to Scan Social Media

Customs and Border Protection (CBP) is using various artificial intelligence tools, including ones that can quickly identify people of interest by pulling data from social media, according to documents published by the agency last week and marketing material from the contractors.

The documents follow others that 404 Media has previously obtained and published through the Freedom of Information Act and are more up to date. 

“Fivecast ONYX supports the collection of publicly available, internet-based content in near real-time, quickly organizing the content to facilitate analysis and the surfacing of risks and threats against the homeland to enhance CBP tactical targeting, analysis and vetting capabilities while at the same time supporting CBP’s strategic counter-network analysis,” one of the documents says, referring to the Australian intelligence company Fivecast. CBP also published documents about Dataminr, another social media monitoring company.

CBP published the latest documents to its website which lists AI programs or tools related to the agency on April 7. Previous documents published in the last year include ones discussing an underwater inspection system, anomaly detection algorithms, and its use of autonomous surveillance towers. 

The release of these documents happened the same week that the Department of Homeland Security (DHS) announced it will “Begin Screening Aliens’ Social Media Activity for Antisemitism.” U.S. Citizenship and Immigration Services (USCIS), a different branch of DHS, is conducting the “antisemitism” social media searches, according to the announcement. CBP told 404 Media in an email that “Neither tool is used for vetting or travel application processing,” referring to Dataminr and Onyx, but did not elaborate beyond that. 

The release of information about these tools comes in a climate of high-profile detainment of students who have criticized Israel’s indiscriminate bombing campaign in Gaza. Hundreds of students have had their visas revoked, many for protesting against Israel, and people who have criticized Israel have been detained on the street or at official appointments with DHS. 

A video on Fivecast’s website says Onyx “combines advanced data collection and AI-enabled analytics to rapidly identify people of interest.” Some of those AI capabilities include object and concept recognition from images and videos; natural language processing that looks for “risky” keywords and phrases; a tool that looks for similar phrases across investigations; and user-trainable logo detector, according to the video. 

As 404 Media previously reported, Onyx also includes sentiment and emotion detection over time.

Although published recently, the Fivecast documents date from May 2023. One of them, a statement of work, says “CBP is proactively collecting, harnessing and applying the power of data, intelligence, and advanced analytics.” This statement of work is similar to some 404 Media obtained from CBP through FOIA requests discussing the agency’s bulk pilot of Fivecast in March 2021 and another from 2022. Journalist Todd Feathers also obtained what appears to be the May 2023 statement of work.

The recently added documents to the AI section of CBP’s website also includes Dataminr, a social media monitoring company that also claims to use AI, although the company has been around for years. “In recent years, Dataminr has integrated Generative AI for real-time event description and launched ReGenAI (Regenerative AI), a breakthrough form of Generative AI that automatically regenerates in real time as events unfold. Dataminr’s AI platform is powered by more than 50 proprietary LLMs and multi-modal foundation models, trained on Dataminr’s 12+ year proprietary event archive,” its website reads.

That statement of work from September 2023 says “OSINT Team analysts, working alongside CBP’s tactical targeting/analytical units, intelligence support cells, and enforcement units, use a commercially available tools and analytical tradecraft methodologies to exploit asymmetric data sources to inform targeting, vetting, and network development and analysis workflows.”

The document adds “DataMinr enables advanced search, collection, and analysis of publicly available information through a single user interface, facilitating the collection of information regarding people, places, and things across social media platforms, as well as general information held on the surface, deep, and dark web to inform situational awareness and to support CBP law enforcement and national security operations.”

I previously obtained a document which said CBP used another AI-powered monitoring tool called Babel X to screen travelers, including U.S. citizens. 

Neither Fivecast nor Dataminr responded to a request for comment.