AI will not replace immigration lawyers. Federal regulation draws the line. Filling in a form is "preparation." Applying legal judgment to a client's facts is "practice." Only a licensed attorney or accredited representative can practice law before immigration agencies. AI tools stay on the preparation side of that line.
This is not just theory. Regulators have disciplined AI products that crossed it. Courts have sanctioned attorneys who let AI write filings unchecked.
USCIS runs AI on its own side of the desk too. That changes what your attorneys need to watch for. Below is what the law, the ethics rules, and the enforcement record say about where AI's role ends and an attorney's begins.
Key Takeaways
- Federal regulation (8 CFR § 1001.1) defines "practice" as applying legal judgment to a client's facts. Only a licensed attorney can do that. AI tools handle "preparation," the mechanical filling of forms.
- The FTC fined an AI legal-services company $193,000 for marketing itself as a substitute for a lawyer. A California lawsuit made the same accusation in court.
- ABA Formal Opinion 512 and Model Rules 1.1, 5.3, and 5.5 require a supervising attorney to verify AI output. They also keep it from crossing into unauthorized practice.
- EOIR Policy Memorandum 25-40 allows generative AI in immigration court filings. But it treats hallucinated citations as a professional conduct violation, against a backlog of more than 3 million pending cases.
- USCIS runs its own AI tools for document translation, identity matching, and fraud screening. That means firms need to understand and challenge algorithmic findings, not just draft the original filing.
- Federal labor data shows lawyer employment still growing. AI is expected to absorb more of the routine work paralegals used to handle by hand.
What AI Handles vs. What Stays With Your Attorney
Some parts of an immigration case are mechanical. Software can run those on its own. Other parts need a licensed attorney's judgment. That is a legal requirement, not just habit.
The right-hand column is not a matter of custom. It comes from a regulatory definition, ethics rules built on that definition, and enforcement actions against companies that ignored it. The next three sections walk through each one.
Where the Law Draws the Line Between Automation and Practice
Immigration has its own regulatory definition of what counts as practicing law. It predates the AI debate by decades. The two sections below cover what that definition says, and what happens to AI products that cross it.
The Legal Definition: Practice vs. Preparation
Immigration regulations define the boundary in exact terms. Under 8 CFR § 1001.1, "practice" means using professional judgment to give legal advice on a matter before the immigration agencies.
"Preparation" is narrower. It means filling in blank spaces on a printed form. The information has to come from the applicant.
That distinction explains a lot. Automation software can pre-fill an I-130 from intake data. That is preparation. It cannot decide which visa category a client should pursue, or how to frame a hardship argument. That is practice, and practice requires a license.
Enforcement Against AI Tools That Cross the Line
Regulators and courts have already enforced that line against AI legal products sold to consumers. The clearest example is DoNotPay, the self-described "robot lawyer."
Neither case involved immigration law. Both show the same principle at work. Software that outputs legal conclusions, not just filled-in forms, is practicing law. Only a licensed attorney can do that.
The Ethics Rules That Keep a Human in Charge
On July 29, 2024, the American Bar Association's Standing Committee on Ethics and Professional Responsibility issued ABA Formal Opinion 512, "Generative Artificial Intelligence Tools." The opinion applies existing Model Rules of Professional Conduct to AI. It does not write new rules.
Three rules matter most for immigration practice:
- Model Rule 1.1 (Competence). A lawyer must understand what a tool can and cannot do. The lawyer must also check its output before relying on it. AI tools can assist a lawyer, but they cannot replace the lawyer's own judgment.
- Model Rule 5.3 (Responsibilities Regarding Nonlawyer Assistance). The duty to supervise nonlawyer assistants covers AI tools too. A supervising attorney owns what the software produces under their name.
- Model Rule 5.5 (Unauthorized Practice of Law). A lawyer cannot let a nonlawyer tool, including AI, perform tasks that amount to practicing law. This rule mirrors the practice-versus-preparation line above.
In practice, this means an automation platform can draft an I-130 petition from intake data. It cannot certify that petition. An attorney has to open it, check it against the facts, and sign off, every time.
AI Model Providers Are Drawing the Same Line
Even the companies that build these models are adopting the same standard. On October 29, 2025, OpenAI updated its usage policies. The new policy bars using its models for "tailored advice that requires a license, such as legal or medical advice, without appropriate involvement by a licensed professional."
That line came from OpenAI itself, not a bar rule or a regulation. The company that builds the model draws the same boundary the bar does.
What Immigration Courts Require of AI-Assisted Filings
Immigration court has its own guidance, on top of the ABA rules above. On August 8, 2025, EOIR Acting Director Sirce E. Owen issued Policy Memorandum PM 25-40, "Use of Generative Artificial Intelligence in EOIR Proceedings."
EOIR's Policy on Generative AI
The memo does not ban generative AI in immigration court filings. It does not require attorneys to disclose that they used it either.
But it draws a firm line on accuracy. A filing with hallucinated or wrong AI content can violate professional conduct rules. It can also expose the attorney to discipline. Individual immigration judges can issue their own standing orders on AI use too.
Why Accuracy Is the Real Risk
The risk behind that warning is not hypothetical. Two facts make it concrete: how often AI hallucinations land in court, and how much backlog pressure pushes firms to lean on AI in the first place.
AI Hallucinations Are Already Landing in Court
Legal researcher Damien Charlotin runs a database that tracks court cases involving AI hallucinations. It had logged over 2,000 cases worldwide as of this writing. Outcomes ranged from formal warnings to monetary sanctions.
The Backlog Is Pushing Firms Toward Faster Tools
More than 3 million cases were pending before the Immigration Court as of the end of August 2026, according to Syracuse University's Transactional Records Access Clearinghouse (TRAC). A backlog that size pushes firms toward faster drafting tools. That is exactly why EOIR's accuracy warning matters.
For a firm, the takeaway is simple. Before anything AI-drafted goes into a filing, a licensed attorney checks every citation, fact, and legal claim by hand. That step is not optional under EOIR's guidance. It is not optional on the government's side of the filing either.
How the Government Uses AI on Its Side of the Case
AI in immigration is not just a law-firm tool. USCIS runs its own AI systems on the same cases your firm files. The agency lists them in its public AI use case inventory.
Three categories are worth knowing:
- Document translation. USCIS uses AI translation tools to convert foreign-language documents. The tools also support real-time interpretation for refugee and case processing.
- Identity and record matching. Systems like the Verification Match Model and the Person-Centric Identity Services Deduplication Model use machine learning to match names, dates of birth, and other identifiers across USCIS databases. They build a single history for each applicant.
- Fraud screening. A text-analytics model scans language patterns in applications, including asylum filings. It flags potential fraud, national security, or safety concerns for further review.
None of these systems issue a final decision on their own. A USCIS officer still adjudicates the case. But an algorithmic match or fraud flag can shape what that officer reviews first. When that happens, an attorney needs to understand the flag, and where warranted, challenge it. That is a judgment call, not a data-entry task.
The Judgment Calls Software Can't Make
Some parts of a case need a human, no matter how good the software gets. Two kinds of work sit permanently on the attorney's side: weighing discretionary legal standards, and the human relationship a case runs on.
Discretionary Legal Standards
Immigration law leans heavily on discretionary legal standards. These require weighing facts against a legal test, not matching a pattern. Proving "extreme hardship" for a waiver takes a persuasive, fact-specific argument. So does establishing a "well-founded fear of persecution" for asylum, or showing a petition serves the "national interest" for a labor-certification waiver.
Software can help gather and organize the underlying evidence. It cannot weigh that evidence the way a skilled advocate can.
Courtroom Advocacy and Client Trust
The same holds in front of a judge. Direct examination, cross-examination, and real-time strategy shifts during a removal hearing all depend on reading a witness, a judge, or a DHS attorney in the moment.
Client counseling adds another layer. Many immigration clients carry trauma, family separation, or fear of a government process. Building the trust that leads to a complete, honest case history is a human skill.
These are the tasks that stay on an attorney's desk. They are also why firms are rethinking how they staff and bill for the work AI does take off their plate.
What This Means for Your Firm's Staffing and Pricing
Everything above is about keeping a human in charge of judgment. That has a practical side effect too: it changes how firms staff and price their work. The shift shows up in two places, federal employment projections and how firms bill.
Lawyer and Paralegal Employment Trends Are Diverging
Federal labor data points to more capacity per attorney, not fewer attorneys. The U.S. Bureau of Labor Statistics projects lawyer employment to grow 4 percent from 2024 to 2034. That is about as fast as the average for all occupations, with roughly 31,500 openings a year.
The picture looks different on the administrative side of a case. The Bureau of Labor Statistics projects little or no employment growth for paralegals and legal assistants over the same decade. It names AI directly as a factor.
The agency expects AI to make research and document preparation faster. That may lower demand for some of the routine tasks paralegals have handled by hand.
The Shift Toward Flat-Fee Pricing
That divide is starting to show up in pricing too. AI compresses the hours spent on drafting and document review. That means hourly billing now rewards slower work over faster, more accurate work.
A growing number of immigration and other high-volume practices are moving toward flat-fee and outcome-based pricing instead. That is a pricing decision for firm owners to weigh on their own. It is separate from the ethics obligations above, which apply no matter how a case is billed.
None of this is a story about lawyers disappearing. It is a story about where the routine work goes. AI absorbs intake, document chasing, form drafting, and translation. Attorneys keep the judgment calls the law and the ethics rules require them to keep. Firms that adopt this kind of software typically handle more active cases with the staff they already have, not a smaller team.
Handle More Cases Without Adding Headcount
US Immigration AI automates the intake, retainer, document collection, drafting, and case assembly work behind every filing. Your attorneys spend their time on legal judgment instead of paperwork. See how the platform fits your firm's workflow.
Frequently Asked Questions
