By Stuart Kerr, Technology Correspondent, LiveAIWire

Algorithmic borders are no longer a metaphor. In 2026, the decision about whether a visa applicant gets flagged for extra scrutiny, whether a document submitted with an immigration petition gets read correctly, and whether a traveler’s face matches someone on a watchlist increasingly runs through machine learning models before it reaches a human officer. U.S. Citizenship and Immigration Services, Customs and Border Protection, and the State Department have all expanded AI use across case processing, fraud detection, and security screening, according to a legal industry alert published in April 2026 by immigration attorneys Scott Bettridge and David Adams of Cozen O’Connor.

The expansion is real and it is measurable in outcomes, not just in press releases. The same alert documents rising Requests for Evidence, Notices of Intent to Deny, and outright denials across multiple visa categories, with EB-2 National Interest Waiver denial rates climbing to nearly 40 percent as AI-assisted screening has expanded.

How Algorithmic Borders Actually Work Inside the System

USCIS now deploys what the Cozen O’Connor alert calls the ELIS Evidence Classifier, a machine learning tool that automatically tags uploaded evidence and determines which documents an adjudicating officer sees first. The system has processed more than 24 million page scrolls, according to the alert, and has measurably changed how officers review filings. USCIS has not published error rate data for the classifier, but immigration attorneys report Requests for Evidence issued for documents that were, in fact, submitted, a pattern consistent with the classifier simply mis-tagging evidence rather than the paperwork actually being missing.

A separate identity-matching system called the Verification Match Model cross-checks names, dates of birth, and other identifiers across databases including E-Verify and SAVE. Minor inconsistencies, a middle name recorded differently across two documents, for instance, can trigger automated flags that lead to delays or erroneous rejections of properly filed petitions. The Cozen O’Connor alert points to one especially stark consequence: more than 1,200 international students recently lost their legal status due to automated database terminations, a problem serious enough to trigger nationwide litigation and a government reversal.

Biometrics at the Border

The screening does not stop at paperwork. CBP now uses a tool called Babel X to analyze social media and other open-source data, including sentiment analysis, to screen travelers at ports of entry, a practice that applies to visa holders and U.S. citizens alike according to the same industry alert. Applicants for many visa categories must now list social media identifiers going back five years, and consular officers use AI tools to flag posts for further review, a process that has extended visa-stamping delays by months in some cases.

Facial recognition has followed a similar expansion path. Procurement records reviewed by Biometric Update show that CBP formally integrated Clearview AI’s facial recognition platform into its National Targeting Center and Border Patrol intelligence units through a February 2026 contract, expanding on smaller 2025 purchases for individual Border Patrol sectors. Clearview’s database draws on more than 60 billion images scraped from the open internet, and the contract requires CBP to complete a formal privacy review before deploying the tool at that scale.

The contract itself frames the technology in narrow operational terms, describing it as a way to support “tactical targeting” and “counter-network analysis” within CBP’s existing intelligence workflows rather than as a new surveillance programme requiring separate congressional authorization. As of the contract’s publication, no Clearview-specific privacy impact assessment had been made public, leaving the deployment in what Biometric Update’s reporting describes as a compliance gray zone: not necessarily unlawful, but not documented in the way federal privacy rules typically require for a biometric tool operating at this scale.

Algorithmic Borders and the Accountability Gap

None of this is happening in a policy vacuum. In February 2026, a group of federal lawmakers introduced legislation that would bar ICE and CBP from acquiring or using facial recognition and other biometric identification technology, and would require deletion of data already collected through those systems. The bill has not passed, and the agencies continue expanding biometric tools in the meantime, but its introduction reflects a documented, bipartisan-adjacent concern among privacy and civil liberties groups that the pace of deployment has outrun the oversight built to govern it.

That concern echoes a pattern this publication has tracked closely in our reporting on AI in law enforcement more broadly, where independent testing has repeatedly found facial recognition systems misidentify some demographic groups at far higher rates than others, and where officers who receive an AI-generated match face real pressure to treat it as confirmation rather than a lead requiring independent verification. The same automation-bias problem now shows up in an immigration system where a mistagged document or a flagged social media post can trigger a visa denial with real consequences for someone’s ability to work, study, or reunite with family.

What This Means for Applicants and Employers

For visa applicants and the employers who sponsor them, the practical guidance from immigration counsel is to expect longer, less predictable processing timelines and to treat social media accounts as part of the application file, since consular AI tools are now reading them as such. Employers should prepare foreign national employees for the likelihood of an RFE even on strong, well-documented cases, since evidence misclassification by USCIS systems has become a documented driver of unnecessary requests.

Immigration counsel are also advising clients to keep independent copies of every document submitted, timestamped and organized, precisely because an automated evidence-classification error is now a foreseeable point of failure rather than a rare exception. That advice would have sounded like excessive caution five years ago. Against a system processing tens of millions of case actions through models whose error rates are not publicly disclosed, it now reads as ordinary due diligence.

The broader response to AI-driven surveillance is not limited to lawsuits and legislation. A parallel ecosystem of privacy-preserving tools, described in our coverage of the growing digital resistance movement, has emerged specifically to help people document and challenge automated decisions that affect them, from freedom of information requests targeting undisclosed AI systems to encrypted communication tools designed to limit what surveillance infrastructure can collect in the first place.

Whether that accountability infrastructure catches up with this system before more people are wrongly flagged, delayed, or denied is, at this point, an open and consequential question. The same oversight gap that has produced wrongful arrests in domestic policing, tracked in our reporting on algorithms used to police the police, is now showing up at ports of entry and consular windows, with the same underlying pattern: automated systems treated as more definitive than their documented error rates justify.

Algorithmic borders are not going away, and the efficiency case for using AI to process millions of immigration cases a year is not unreasonable on its face. What remains unresolved is whether the systems making those decisions will be held to the same disclosure, error-rate, and human-oversight standards that any other high-stakes government decision-making process would be expected to meet.

About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, emerging technology, and their impact on business, society, and everyday life. LiveAIWire publishes original AI journalism every weekday at liveaiwire.com.