How AI Is Reshaping US Immigration Strategy and Compliance

Grace Shie: Welcome to The Inside Track. My name is Grace Shie and I'm joined today by my partner Morgan Bailey. In our recent episodes, we have examined the latest policy announcements or enforcement actions that's made front page news. And today Morgan and I will examine something more stealth, something quietly but steadily operating in the back and that is the role of AI in the US immigration system. We're seeing AI used in so many different areas of our lives, whether professional or personal, and certainly both in the public and private sectors. Now we are seeing the gradual integration of AI into immigration agency workflows.

Morgan Bailey: That's right, Grace, and it's happening quietly, not through sweeping regulatory announcements, but through incremental operational changes.

Grace Shie: And those changes, Morgan, may not be obvious to the average person, but you have a unique perspective on this given your tenure in government service, including at US Citizenship and Immigration Services or USCIS. So I thought today you might first introduce how the various agencies from the Departments of State to Labor to Homeland Security, how they're incorporating AI into their workflows.

Morgan Bailey: The system is beginning to rely more on automation to organize information as well as to detect patterns and to, you know, really to support decision making by immigration officers. So AI isn't just being piloted, it's actually being used. AI is not necessarily replacing officers or adjudicators. In many instances, the ultimate decisions are still made by people. But AI is increasingly influencing what information reaches them and how that information is presented and really where officer attention is directed. It's also a system that's evolving through repeated use and modernization. While the use of AI is frequently framed as efficiency, which is an accurate description. Some adjudicators or adjudications are moving much faster as a result, while others are taking longer and in some instances facing more detailed requests for evidence or clarification. It's not just about speed, rather there is really a shift towards more continuous analysis of filings. So that means ongoing scrutiny across the life cycle of a case.

Grace Shie: So when we see variances in adjudication, some faster, some slower, as you mentioned, or cases receiving different outcomes, despite similar or the same or identical fact patterns, oftentimes practitioners, we conclude that those variances relate to the different officers who are assigned to review those cases. But if AI is being used, what you're suggesting is that there's more to it than what we're concluding.

Morgan Bailey: Exactly. The variances aren't random and not necessarily due to having different officers assigned to each case. The variances can may align with hallmarks of AI assisted workflow and pattern recognition. You know, this can also be an anomaly detection and prioritization based on alignment rather than only focusing on eligibility.

Grace Shie: So let's start by sharing some examples within DHS or Homeland Security, specifically the components USCIS, CBP, and ICE. So starting with USCIS, your former agency, we know USCIS is responsible for administering the nation's lawful immigration system. And this includes everything from conducting background checks and fraud prevention checks to adjudicating and processing immigration benefits requests. How do you see AI's use within this agency?

Morgan Bailey: USCIS is a notable example. It's an agency that's really built around evidence, such as government forms and documentary evidence, which might include civil records, such as birth certificates or employment verification, financial evidence, such as tax returns and pay records, academic credentials, immigration history. As well as things like affidavits or letters from others. AI is really well suited to help categorize that information. It can compare filings against past submissions and flag areas where the facts maybe don't appear to fully reconcile.

Grace Shie: Well, immigration filings, as you and I know, are notoriously document heavy. And you've just identified just a short list, frankly, of all the different types of documents that can go into a filing. So an automated means to categorize reams of paper and hundreds, if not thousands of data points, that sounds significant. Can you provide an example that might bring this to life?

Morgan Bailey: Sure, USCIS is using machine learning tools to support fraud detection. So DHS recently announced a new centralized vetting center that's designed to enhance national security screening across immigration systems using really advanced technologies such as AI in coordination with law enforcement and in coordination with intelligence agencies resources. This signals a shift. So rather than handling vetting on a piecemeal basis, the government is consolidating the screening under a high capacity hub with a goal really of identifying fraud, security risks, and also ineligibilities. This scheme may include, for example, scanning written narratives to detect when the same language appears across many unrelated filings by different people. That kind of repetition can signal scripted or mass produced claims rather than individualized accounts. Another example is analyzing the structured data behind petitions. This is things like the preparers or the addresses sometimes timeline patterns or employer links, the tool identifies clusters or repeat submissions that may point to coordinated fraud networks. So essentially, we're looking at tools that analyze the language of the story and tools that can analyze the data behind the filings. When DHS ordered the review of thousands of immigration cases after the National Guard shooting, it was not just intending for there to be a manual file pull. Rather, DHS is using automated screening tools and algorithms to surface risk criteria and to determine record matching. And, you know, AI is the mechanism that makes that mass review possible. The value here isn't just automation, it's really scale. These tools will help officers see patterns that would be really impossible to notice if they were reviewing one case at a time.

Grace Shie: So those are some really powerful examples you've just identified and with probably incredible capabilities given the creation of this vetting center. I guess taking that example to the employment-based context, I could see, for example, an H-1B visa holder having multiple successive H-1B petitions filed on her behalf across multiple years, a career here in the United States. And I could imagine if all but one of the petitions uses the same occupational classification, the one petition with a deviation will now take on greater meaning because it has a higher likelihood of being detected out of the many petitions filed over many years. And if detected upon or through AI without any explanation by the employer, this can be problematic for the H-1B worker and potentially for the H-1B employer. Would you agree?

Morgan Bailey: I think you're exactly right. know, applicants are already seeing the impact. Immigration filings that tell a clear and consistent story often progress more predictably. Meanwhile, filings with really unexplained changes, whether it's a job title or career progression or role responsibilities, they sometimes experience follow-up requests, even when eligibility is strong. And that mirror is really something that DHS has publicly stated. There's a move toward a person-centric model. So rather than treating each petition or forum as a standalone event, the system is beginning to view the applicant as a unified record across time. This means that the immigration history is becoming more cumulative rather than episodic.

Grace Shie: So having just provided some examples of how USCIS uses AI, let's now turn to Customs and Border Protection or CBP. CBP is responsible for safeguarding the US borders, and this means inspecting travelers entering the United States. At the various ports of entry, we already see AI used in quite a visible way. Biometric identity verification, for example, is already standard in many locations. This includes photos and fingerprints captured at the port of entry. And in fact, we have some of these biometrics captured even for domestic travel within the United States. What do you see with respect to CBP and its use of AI?

Morgan Bailey: Officers are now supported by systems that present really a consolidated snapshot of a traveler's history more quickly than ever before. I can think of, know, there's a new rule that was recently published where CBP will begin requiring photographs, also known as facial recognition data as well as other necessary biometrics, but they'll be required from all non-U.S. citizens upon not just the entry to the United States, but also the departure from the United States. This represents a nationwide entry and exit system. It will provide a more complete travel history that can be really continuously analyzed. And the practical effect is clear. Routine travel patterns will be processed faster and in some instances with fewer questions for the individual travelers. But when something doesn't align, such as maybe a name variation or a potential travel inconsistency or a new visa status without context, the system often will immediately flag it for the officer. The traveler is then referred to secondary inspection.

Grace Shie: With secondary inspection being a more detailed review or screening conducted by CBP in typically a secondary room. Is that right?

Morgan Bailey: That's right. And the officers take additional time to verify any duty, confirm citizenship and immigration status, and just to resolve any discrepancies before making a decision whether that person may lawfully enter the United States.

Grace Shie: I think the CBP's use of AI through biometrics and as you say, facial recognition, I think that will be commonly understood for anyone who engages in air travel in particular. So now let's turn our attention to ICE. ICE or Immigration and Customs Enforcement is the agency responsible for enforcing our immigration laws. And the past several months, we've seen their focused prioritization of detention and removal of certain foreign nationals.

Morgan Bailey: Yes, and on the enforcement side, the trend really appears to be towards more targeted actions, such as those involving investigations and site visits or raids, really relating to worker verification rather than just broad sweeps. And AI can highlight where multiple indicators of compliance risk appear. So AI is not replacing ISIS investigations, but it is, it's being used in some instances to identify patterns and to help prioritize leads, as well as to guide where investigations begin. In some cases, we're seeing that it's supporting a deeper analysis on ongoing cases by making connections or identifying anomalies. ICE is really using AI so that investigations are less reactive and more directed towards enforcement.

Grace Shie: And if that really underscores the overall mandate for ICE. So turning away from Homeland Security, let's now look at the U.S. State Department, which is responsible for adjudicating and processing visa applications filed by foreign nationals at our overseas embassies and consulates. We know that the State Department is instructed by executive order to engage in what's known as enhanced vetting. And certainly the use of AI assists in carrying out that mission. Applicants are increasingly reporting on visa interviews that reflect an officer's awareness of prior visa filings or certain prior history involving travel or employer relationships, information or facts that might not have been presented at the window during the visa interview, which speaks to the availability and the search and the unearthing of these facts, this background through the use of AI.

Morgan Bailey: That's right. For example, earlier this year, the Department of State announced that more than thousands of student visas had been revoked. And this was reportedly after an AI-supported review that included student records, prior visa filings, travel history, social media activity, as well as law enforcement records. We're hearing reports that the Department of State is now focusing on H1B and other workers for a similar review. So the technology will flag patterns or issues that would have been extremely resource and time intensive to identify on an individual basis. This is another example of the government focusing on that person centric model where the individual is judged over really connected records over years of activity.

Grace Shie: And someone who has applied for multiple visas over many years through the various embassies. So under the State Department's jurisdiction, that means, as you said, those applications are not viewed as standalone episodes now. It's a cumulative history. So let's not forget the Department of Labor, particularly the Labor Department's role in wage determinations that are used in the H-1B context, for example, as well as in perm application processing for green card sponsorship. We know the Department of Labor is exploring AI automation because of the volume and rule-based nature of that adjudications process, and we're already seeing elements of that shift.

Morgan Bailey: The rollout for AI for all of these agencies is really still evolving as they work through the implementation, accuracy, and really the operational realities.

Grace Shie: So with that review of how AI is being used by these various departments, let's discuss what this means for organizations and individuals who are navigating the system today and the themes that are emerging.

Morgan Bailey: First, as an initial matter, it really requires a shift in mindset. So immigration cannot be treated just as a series of independent one-time transactions. The immigration process is really becoming more holistic, drawing on a fuller picture of a person's history. The immigration process and related issues such as, let's say, an individual's background or any criminal issues, the payment of taxes, all of this is now more intertwined and cumulative. The submissions really build upon one another. And when the story aligns across time, the process tends to move more smoothly.

Grace Shie: That's right. And to tag on to that, we've discussed how an explanation, a real time explanation, a contemporaneous explanation is more valuable than a correction that takes place in the future. If something changes, change of a job title, change of an entity or employer, or a change in the travel pattern or rhythm, a brief but clear explanation at that time may prevent the system from treating that particular variance or deviation in the future as an unresolved question and potential inconsistency treatment as an inconsistency.

Morgan Bailey: And along those same lines, the neutral gaps are not always neutral anymore. So silence can look like missing data rather than simplicity. So if we're thinking like the system, if a career path looks nonlinear or a business structure has evolved, it really is important to frame it so it reads as intentional rather than irregular. And a key perspective shift is that patterns matter as much as eligibility. You can fully qualify and still create friction if the record feels disjointed to the reviewing officer.

Grace Shie: And I'll add that documentation hygiene matters, we just discussed, you know, immigration filings built over a lifetime can be hundreds of thousands of pages. And we talked about hundreds and thousands of data points, but certainly also hundreds and thousands of pages. And the use of AI means that variants across the reams of paper can be detected quite easily and quite quickly. So that means clarity, labeling of the documentation that serves as evidence and structure, all of that together can reduce ambiguity in your profile. And I would say that this is true for both an automated review and also a review by officers where there is a live adjudication. So with that, any final advice, Morgan, on this topic?

Morgan Bailey: Well, I think just that, you know, AI can misread context. It can misunderstand nuance or surface patterns that look meaningful, but really are not. So that aspect of transparency and human judgment really remain critical. We're not describing a system that's fully already transformed. Rather, I think we're describing one that's actively shifting and it's not flawless. And those who recognize that early will be able to navigate it with less friction and more predictability.

Grace Shie: Thank you for sharing those insights, Morgan. Here at Mayer Brown, we will continue to watch how these developments unfold. The use of AI is here to stay, and we are likely only starting to understand and to experience the system's capabilities and how those capabilities will be used by the new screening and vetting center. That wraps up this episode of The Inside Track. Thank you for joining us.