Documents Detail DHS Project to Give ‘Risk Scores’ to Social Media Users

The Department of Homeland Security contracted the University of Alabama at Birmingham (UAB) in 2018 to design methods for assigning a “risk score” to potential pro-terrorists accounts on social media, as well as identifying information of interest regarding illegal opioid supply chain and disinformation efforts, according to internal DHS documents reviewed by Motherboard. The project is dubbed “Night Fury,” according to a report from the DHS Inspector General.

“The Contractor shall develop these attributes to create a methodology for developing a ranking, or ‘Risk Score,’ associated with the identified accounts. The Contractor shall develop tools to automate the identification process, documenting performance measures and metrics related to automating the identification process,” one of the documents reads. DHS said it stopped work on the project in 2019.

The news signals DHS’ continued focus on analyzing social media for a variety of purposes. These new documents come after Motherboard reported Customs and Border Protection (CBP) was using an AI-powered tool called Babel X to analyze travelers’ social media at the U.S. border.

The Brennan Center for Justice obtained the new documents under a public records request and shared them with Motherboard. They include Privacy Threshold Analyses of the project and contracts. The research planned to involve CBP, ICE, TSA, and USCIS which would provide “cross-mission operational context,” one document reads.

“The use of automated processes to analyze social media to determine the likelihood that someone is ‘pro-terrorist’ and to assign a ‘risk score’ to individuals and groups online has echoes of a discredited Trump administration proposal called the Extreme Vetting Initiative, which would have monitored social media and the rest of the open internet to automatically flag people for deportation or visa denial based on whether they would be a ‘positively contributing member of society’ or ‘contribute to the national interests,’ as well as whether they ‘intend to commit’ a crime or act of terrorism,’ Rachel Levinson-Waldman, Managing Director, Liberty & National Security Program at the Brennan Center for Justice, told Motherboard in an email.

“As a number of experts in machine learning and automated decision-making told DHS less than a year before the Night Fury contract was signed, attempting to make automated judgments about these matters is both impossible and likely to be infected with bias, as these characteristics have no concrete definition, much as there is no definition of being ‘pro-terrorist,’” she added.

{snip}

More specifically, the project planned to develop methods that could identify a location without GPS metadata, such as looking for certain keywords, the document reads. The researcher also planned to track threats beyond mainstream social networks like Facebook and Twitter to other communities. {snip}

Another task was to create a “Facebook Group Expander,” which would automatically identify potential pro-terrorist social media accounts and Facebook Groups where pro-terrorist groups interact, one document reads. {snip}

{snip}