Using AI to Cancel Visas Based on Social Media Posts Endangers Democratic Society
This so-called “anti-terror” measure is being used to terrorize foreigners and to dissuade people from participating in First Amendment-protected activity for fear that they too will be targeted in some way.
On March 8, U.S. Immigration and Customs Enforcement, or ICE, agents arrested Mahmoud Khalil, a Columbia University student and a prominent leader of pro-Palestinian protests on the university’s campus. They claimed that Khalil’s student visa had been revoked and, when told that he had a green card, said that too had been revoked.
While the full facts of the case are yet to emerge, there seems little doubt that Khalil was detained in retaliation for his activism. U.S. President Donald Trump has frequently and explicitly threatened to go after university protestors, including in his Executive Order on “Protecting the United States from Foreign Terrorists and Other National Security and Public Safety Threats,” which I analyzed in an earlier post. Trump celebrated Khalil’s arrest on social media, warning that it was the first “of many to come.”
Some of the “many to come” will likely be identified via the State Department’s newly launched AI-enabled “Catch and Revoke” initiative, which will scrape social media to find “foreign nationals who appear to support Hamas or other designated terror groups” and cancel their visas. Like the executive order cited above, this effort is framed as an anti-terrorism measure. Instead, it is being used to terrorize foreigners and to dissuade people from participating in First Amendment-protected activity for fear that they too will be targeted in some way.
It is part of Trump’s broader effort to subdue all potential sources of opposition by attacking universities, the press, law firms, and jurisdictions that refuse to cooperate with ICE.
Starting with the Obama administration, the federal government has built an extensive infrastructure for agencies to comb social media looking for certain types of speech. Even as civil society groups have raised concerns about how these programs could be used to target unpopular speech, they have continued to proliferate.
The State Department, for example, collects social media handles from certain types of visa applicants—some 14 million people a year—which are saved indefinitely in government databases. (The Brennan Center, where I work, and the Knight Institute have challenged this program in court.)
The second Trump administration is aiming to dramatically expand these efforts, collecting social media identifiers from an additional 33 million people, including those applying for permanent residence or adjustment of their immigration status. The first Trump administration’s attempt to do so was blocked in 2021 by the Biden White House’s Office of Information and Regulatory Affairs on the grounds that the government had not demonstrated “the practical utility of collecting this information.”
The U.S. Department of Homeland Security (DHS) runs at least 12 overlapping programs that track what Americans say online, several of which are focused on protests. DHS used social media to track protests against the first Trump administration’s immigration policies. During the Biden administration, DHS scanned social media for other targets, such as Americans discussing abortion after the Supreme Court overturned Roe v. Wade and trucker convoys protesting Covid-19 mandates, as well as broadly monitoring online “narratives and grievances”—i.e., people talking politics.
Even as it adds more and more social media monitoring programs to its repertoire, the government has never shown that these efforts are effective. A February 2017 DHS Inspector General audit of six pilot programs found that the department had not even measured their effectiveness. And the few government evaluations that are publicly available undermine any governmental claims of efficacy. A brief prepared by DHS for the first Trump administration concluded that social media monitoring did not provide useful information for vetting refugees. And, according to a 2021 analysis by the Office of the Director of National Intelligence, social media identifiers added “no value” to the immigration screening and vetting process.
Looking for unknown foreign protestors who may have made ostensibly pro-terrorist statements is much harder than vetting the posts of a known person, such as a visa applicant. It will undoubtedly sweep far too broadly and result in mistakes. The AI tools that will be deployed by the State Department likely will be tasked to search for specific words or phrases. The Trump administration has used these types of lists in its attempt to root out diversity, equity, and inclusion programs in the federal government, resulting in various blunders. In one instance, a federal employee who managed relations with private equity-held businesses was placed on administrative leave “pursuant to the president’s executive order on DEIA.” The Internal Revenue Service purged its employee manual of references to the “inequity” of holding on to taxpayer money longer than necessary and the “inclusion” of a taxpayer identification number on a form. The Defense Department flagged for deletion mentions of the World War II Enola Gay aircraft and references to people who have the last name “Gay.”
Even without mistakes, broad social media monitoring will have enormous First Amendment consequences. The types of speech that the administration has declared it intends to target is exceptionally broad. In defending his arrest, DHS said Khalil led activities “aligned” with Hamas, a term untethered to any law or regulation. Statements from Trump and his cabinet characterize foreigners who are in the administration’s crosshairs as “pro-Hamas” (most common), “pro-terrorist,” “terrorist sympathizers,” people who “support terrorism,” and “antisemitic.” These are broad and contested terms. Pro-Palestinian and anti-Israel sentiments have often been conflated with antisemitism or pro-terrorism, leaving a broad swath of people vulnerable to being caught in an AI-enabled social media net.
The Trump administration’s efforts ultimately may sweep even more broadly, seeking out speech that it views as anti-American. The vetting executive order instructed the Secretary of State to recommend measures for foreign nationals who call for the “overthrow or replacement of the culture on which our constitutional Republic stands.” In addition, the sole justification provided by the administration for acting against Khalil is a single line in the Notice to Appear in immigration court: “The Secretary of State has determined that your presence or activities in the United States would have serious adverse policy consequences for the foreign policy of the United States,” citing 237(A)(4)(c)(1) of the Immigration and Nationality Act.” As Adam Cox and Ahilan Arulanantham explained on Just Security, this provision cannot be read as a blank check for the administration to deport people based on an unarticulated foreign policy rationale. But if the administration wants to deport foreigners who take positions contrary to U.S. foreign policy, they will certainly find plenty of fodder on social media.
Khalil’s case and the Trump administration’s promise to go after foreign protestors for their social media posts is an extraordinary assertion of executive power over immigrants living in the United States. But it should not be viewed in isolation. It is part of Trump’s broader effort to subdue all potential sources of opposition by attacking universities, the press, law firms, and jurisdictions that refuse to cooperate with ICE. All of these endanger the fundamental constitutional promise of a democratic society in which a multitude of views and interests can be freely expressed.
