For most of the US immigration law's history, adjudication meant a human being reading your file and making a decision. That human could be wrong, could be biased, could have a bad day. But there was a person there. You could argue with their reasoning. You could appeal it. You could at least identify it.
That model is changing. Artificial intelligence is now embedded in the immigration adjudication process at multiple levels, and the effects are no longer theoretical. RFEs have surged. Thousands of international students, the majority of them Indian, had their legal status terminated overnight by a system relying on a database that does not even track whether criminal charges were ever pursued. Courts intervened repeatedly. The government was embarrassed repeatedly. And yet the machine continues to expand its reach.
This article is about what is actually happening, what it means for you, and what you need to do about it.
What USCIS Is Actually Deploying
Let us start with the facts DHS has publicly acknowledged, because the government's own disclosures are more alarming than most reporting has captured.
The DHS Artificial Intelligence Use Case Inventory lists approximately 29 USCIS-specific AI use cases as of its February 2025 update. Across all of DHS, over 200 AI systems are in active deployment, many of which remain undisclosed to the public. These are not pilot programs. They are operational tools running at scale.
The most significant USCIS AI tool for everyday petitioners is the ELIS Evidence Classifier, a machine learning system that automatically tags and categorizes every document you upload with your petition. USCIS says it has saved approximately 24 million page scrolls since its launch. That sounds efficient. The problem is that this system also standardizes which evidence gets surfaced to adjudicators and which gets deprioritized. If your documents are not clearly labeled, logically organized, and neatly matched to expected evidence categories, the system may not present them optimally. A human adjudicator reading a file can recognize what something is, even when it is presented imperfectly. An Evidence Classifier may not.
USCIS also deploys the FDNS-DS NexGen system through its Fraud Detection and National Security Directorate. This uses AI to prioritize fraud investigation cases, identify individuals who may pose national security concerns, and flag anomalies across large datasets. A Text Analytics Sentence Similarity Model analyzes text in asylum and refugee contexts. An Automated Name and date-of-birth harvesting tool pulls aliases and identity data from FBI background check systems. A workload forecasting model predicts filing volumes. There is no public indication that it is currently used in employment-based cases, though the underlying technology could readily be extended to analyze patterns across high volumes of employment-based filings in the future.
DHS insists that human officers retain final decision-making authority. That is technically accurate. But the environment in which a human officer makes that decision is now significantly shaped by what an algorithm has already flagged, sorted, and surfaced before the officer ever opens the file. The question is not whether AI makes the final call. The question is how much the algorithm has already pre-determined what the officer sees, and in what order, and with what flags attached.
There is no publicly available data comparing approval, denial, and RFE rates in AI-assisted adjudications versus traditional adjudications. DHS has not published it. That absence of transparency is itself a problem.
The RFE Surge: Numbers and Causes
Requests for Evidence have increased sharply since January 2025, and the trend is consistent across nearly every employment-based petition category.
One immigration firm tracking its own case portfolio reported a 23 percentage-point increase in H-1B RFE rates in the first months of 2025 compared to 2024. Historical data from the first Trump administration is instructive: RFE rates under Trump 2017 to 2020 ran 20 to 40 percentage points higher on average than Biden-era rates. We are tracking toward, and likely beyond, the peaks of that earlier period.
The data across petition categories confirms the pattern. NIW approval rates have dropped 13 percentage points, falling to 54% as of Q3 2025. EB-1A approval rates have dropped to 67%, the lowest in three years. The EB-1A backlog grew 67% in a single quarter, largely because surging RFEs keep cases open for months while petitioners respond. O-1 approvals remain above 90%, but RFEs in that category are described by practitioners as broader and more demanding than before.
In our own practice, we have seen this across all categories. Petitions that would have sailed through with the same documentation two years ago are coming back with detailed evidence requests. Some of the RFEs we are receiving on NIW cases question the "national importance" of work in fields that USCIS's own Policy Manual would seem to clearly qualify. The discretion being exercised is real, and it is being exercised aggressively.
There is an additional and subtler phenomenon worth noting. Some practitioners have reported receiving RFEs on concurrent adjustment filings on the same day receipt notices were issued. In the documented instances, the RFEs challenged Form I-864 data with incorrect deficiency claims, consistent with automated parsing of financial support documentation at intake. Think about what that means. A petition arrives. A receipt is generated. An RFE is issued the same day. No human adjudicator reviewed the file in that window. What is more likely: an adjudicator worked that fast, or automated intake screening flagged something and auto-generated a deficiency notice before a human ever looked at the case? USCIS has not confirmed this. But the pattern is consistent with how the Evidence Classifier and intake screening tools are documented to operate.
To put matters in context, the surge in RFEs has both a policy cause and a probable technological one. The policy cause is clear: the current administration has explicitly instructed USCIS officers to apply stricter scrutiny and has given officers expanded authority to deny without RFE when documentation is incomplete. The technological cause is more opaque, but the timing of same-day RFEs and the consistency of certain deficiency patterns across unrelated cases suggests that automated tools are shaping, and in some cases triggering, the process.
Indians are among the most heavily affected. Indian nationals file the largest volume of H-1B petitions by far, the largest share of EB-1A and NIW cases, and a substantial portion of O-1 filings. When RFE rates climb across all of these categories simultaneously, Indian professionals bear a disproportionate share of the burden, both in additional legal costs and in extended delays for cases that feed into an already years-long green card backlog.
The SEVIS Catastrophe: What Happened and Why It Matters
The RFE surge is an inconvenience, sometimes a serious one. What happened to international students beginning in March 2025 was a different order of magnitude. It was one of the most legally indefensible mass immigration actions in recent memory, and the courts said so.
Beginning in late March 2025, the Department of Homeland Security began terminating SEVIS records for international students on F-1 visas, mostly without prior notice and without individualized explanation. By May 7, 2025, over 4,700 students had their records terminated. SEVIS, the Student and Exchange Visitor Information System, is the federal database that tracks the legal status of international students. When your SEVIS record is terminated, your ability to study, your work authorization if you are on OPT, and the legal status of any family members on derivative visas are all immediately at risk.
The government ran these terminations using the National Crime Information Center database, a law enforcement records system. The core problem is that the NCIC does not systematically track case outcomes. It records encounters and charges. It does not reliably reflect whether those charges were dismissed, whether the student was ever convicted of anything, or even whether the student was the correct person in the record. Students were terminated for parking tickets. For minor speeding citations that had been dismissed. For interactions with police in which no charges were filed. In several cases, students had never had any contact with law enforcement whatsoever. Only two of the documented cases involved political activism.
According to data reviewed by the American Immigration Lawyers Association, approximately half of the students affected were Indian nationals. Chinese students were the next-largest group, at about 14%. Indian students represent the largest single cohort of international students in the United States, are concentrated in STEM fields, and bear the brunt of an action the government cannot legally defend.
Students fought back. At least 65 lawsuits were filed across the country. Courts granted temporary restraining orders in more than 35 of those cases, requiring the government to immediately restore students to active SEVIS status. In case after case, judges found that forcing students to halt their studies mid-semester, lose their work authorization, and face the prospect of deportation constituted irreparable harm.
In one case before the US District Court for the District of Columbia, Judge Ana Reyes asked a government lawyer repeatedly whether terminating a student's SEVIS record meant the student was no longer legally in the country. The government lawyer said he did not know. Judge Reyes responded: "Do you realize that this is Kafkaesque?" She continued: "I've got two experienced immigration lawyers on behalf of a client who is months away from graduation, who has done nothing wrong, who has been terminated from a system that you all keep telling me has no effect on his immigration status, although that clearly is BS."
Faced with overwhelming losses in court, the government reversed the terminations in late April 2025. That reversal was framed as voluntary in some court filings, but it followed a string of judicial orders making the government's continued position untenable.
The reversal was not the end of the story. The administration subsequently announced a new policy framework for SEVIS terminations that would allow future terminations on broader grounds, including visa revocations by the State Department. The details of how that new framework operates are still being contested in ongoing litigation. Students whose records were restored and whose visas were separately revoked found themselves in a legal limbo: SEVIS active, visa revoked, unable to re-enter the United States if they left.
The SEVIS debacle illustrates something important beyond the specific legal questions it raised. It demonstrates what happens when government agencies use automated or semi-automated processes to take mass, adverse action against large populations of people without individualized review, notice, or a legally defensible evidentiary basis. The NCIC database was not designed for immigration enforcement decisions. Using it as a trigger for automatic status terminations was, as the courts repeatedly found, not in accordance with law.
Social Media Screening: The Expanding Surveillance Architecture
Layered on top of all of this is a rapidly expanding social media screening regime that operates largely without published standards, without meaningful notice to applicants, and with no clear appeals mechanism.
On April 9, 2025, DHS formally announced that USCIS would begin incorporating social media activity into its adjudications for F and J visa holders and their family members. The stated focus was on content supporting antisemitic terrorism and antisemitic organizations. But the actual policy language is broader. USCIS is directed to consider any social media content reflecting negatively on an applicant's character or immigration intent as a factor in discretionary adjudications.
The State Department began requiring social media screening for all F, M, and J visa applicants in June 2025. As of December 2025, H-1B workers and their H-4 dependents are subject to the same screening at consular posts. By March 2026, the policy was extended to 14 additional visa categories, including fiancé visas, religious worker visas, and others.
USCIS has been independently vetting applicants' social media accounts, including LinkedIn, Instagram, and other platforms, during the adjudication of change-of-status applications, OPT, STEM OPT, and other benefit requests.
Note that there is currently no published standard explaining what content will trigger an adverse decision. Consular officers and USCIS adjudicators have broad discretion in how they interpret what they find. Past posts, shares, likes, and even group affiliations are all potentially reviewable. Accounts that are not publicly viewable may themselves be treated as a concern at the consular level. Failing to disclose a social media account on Form DS-160, even an inactive one, can raise credibility issues that affect your application.
This is the environment in which employment-based petitions, student visas, and green card applications are now being evaluated. The process is more opaque, more automated, and more expansive in what it considers reviewable than at any prior point in recent history.
What This Means Practically, and What You Should Do
The picture I have described is not one in which AI has taken over immigration adjudications. USCIS officers are still reviewing files and making decisions. But AI is now meaningfully shaping which files get flagged, how evidence is organized and surfaced, what background information appears before an officer reviews your case, and what your social media history may contribute to the analysis. That is a significant shift, and it requires a significant adjustment in how you approach filing.
For employment-based petitions, H-1B, O-1, EB-1A, EB-1B, NIW, and others, the implications are these:
Document organization is no longer just a courtesy to the adjudicator. It is a technical requirement. The Evidence Classifier will categorize what you submit. Clearly labeled, logically structured, purposefully organized evidence is not just more persuasive. It is more likely to be correctly surfaced to the human adjudicator who ultimately signs the decision.
File stronger petitions than you think you need to. The 2025 policy environment, combined with automated intake screening, does not reward the minimum sufficient case. It rewards the comprehensive, well-argued, evidence-heavy case that leaves no obvious gaps for a flag to attach to.
Anticipate the RFE. If you are filing an NIW, think through every weakness in the national importance argument and address it in the initial submission. If you are filing an EB-1A, address every criterion you are not claiming and explain why the presented criteria satisfy the totality standard. The 2025 guidance explicitly allows denial without RFE where documentation is incomplete. You want to make it difficult, not easy, for a gap to become a denial.
For students on F-1 visas, the SEVIS experience offers several lessons. Your legal status in this country and the database record that reflects it are not the same thing. If your SEVIS record is terminated without notice, do not assume you must leave immediately. Consult an immigration attorney before taking any action, including self-departure. Courts have repeatedly found that students have the right to challenge these terminations, and many have succeeded. Leaving voluntarily before that process plays out may forfeit rights you actually have.
Be mindful of your social media presence in a way you may never have had to be before. This is not about self-censorship. It is about understanding that posts, likes, shares, and group affiliations that you may have considered private expressions of opinion are now potentially part of your immigration record. On every platform, in every language.
If you are an H-1B worker or an H-4 dependent going to a consular post for a visa stamp, the same social media screening that applies to students now applies to you. Ensure all platforms are disclosed on your DS-160. Ensure your accounts are publicly viewable before your interview, or be prepared for adverse adjudication.
As a practical matter, the most important thing you can do in the current environment is work with experienced immigration counsel before you file, not after you receive an RFE. The cost of prevention is far lower than the cost of remediation, and in this environment, remediation is neither fast nor guaranteed.
A Final Thought
What the SEVIS debacle demonstrated, and what the RFE surge and expanding AI tools reinforce, is that the government is capable of taking consequential action against large numbers of people based on automated processes that are neither transparent nor legally defensible on their own terms. The courts have corrected some of this. They will continue to be an important check. But litigation is expensive, time-consuming, and accessible mainly to those who can afford it.
The best protection is a filing so thorough, so well-documented, and so clearly organized that even an imperfect algorithm has trouble finding a hook on which to hang a deficiency. The second-best protection is knowing your rights and being prepared to assert them.
The machine is in the room. It does not have the last word. But it has a significant first word. Prepare accordingly.
Rajiv S. Khanna is the Managing Attorney at Law Offices of Rajiv S. Khanna, PC. The firm has been practicing US immigration law since 1993. For questions specific to your situation, please consult with a qualified immigration attorney.