An asylum applicant sits across your desk. She describes what happened in her home country. The details are specific. The fear is credible. You take notes, collect documents, and open a new matter file.
That file now contains her A-number, passport data, biometric appointment records, a detailed persecution narrative naming her persecutors, her home address abroad, family member identities, and possibly evidence of political activity that would get her killed if disclosed to the wrong government.
No other area of legal practice carries this weight. A data breach in a corporate transaction costs money. A data breach in an asylum case can cost a life.
You already know AI would transform your practice. You have seen what it can do with document preparation, legal research, and case management. But you cannot use it. Not with these files. Not on someone else's servers.
That calculus changes when the AI never leaves your office.
Why immigration attorneys are not using AI
It is not ignorance. Immigration attorneys read the same articles about AI-assisted legal research that every other practice area reads. They attend the same CLEs. They see solo practitioners in other fields using AI to draft motions, summarize depositions, and manage caseloads.
They also see what is in their own files.
An I-589 asylum application contains the applicant's complete persecution narrative. Names of family members. Membership in targeted political, religious, or social groups. Specific incidents of harm. Dates, locations, and perpetrators. This is the most sensitive personal information any attorney in any practice area will ever hold.
An I-130 family petition file contains passport scans, birth certificates, marriage certificates, financial affidavits, and tax returns for both the petitioner and the beneficiary. An I-485 adjustment file adds medical examination records, vaccination history, and detailed immigration history including any prior unlawful presence.
A VAWA self-petition contains evidence of domestic abuse. A U-visa application contains a law enforcement certification describing the crime committed against the applicant. A T-visa file documents human trafficking.
The risk calculus is different here. When a corporate attorney's data leaks, the client faces financial exposure. When an immigration attorney's data leaks, the client faces deportation to a country where the government already wants them dead. Or a trafficking network learns their victim cooperated with law enforcement.
In February 2026, Anthropic confirmed that state-linked Chinese AI laboratories extracted 16 million queries from Claude through 24,000 fraudulent accounts. The operation ran undetected for months. The breach vector was the cloud connection.
Now imagine an asylum applicant's persecution narrative in that dataset. Imagine a foreign intelligence service with access to the names, addresses, and political affiliations of dissidents who fled their country and sought protection in the United States.
This is why immigration attorneys are not using AI. The risk is not malpractice. The risk is complicity in harm to the people they swore to protect.
What private AI makes possible
Dhakma Core is a self-contained AI server that sits in your office. Pre-configured. No IT department required. Dhakma installs it, supports it, and keeps it running. Your team accesses it through a secure internal web interface. No data leaves the building. No cloud connection exists. A physical air-gap switch ensures that no software vulnerability can create one.
Here is what that unlocks for an immigration practice.
Country conditions research
The problem: An asylum case lives or dies on country conditions evidence. The immigration judge wants sourced, current, comprehensive documentation of conditions in the applicant's home country. State Department human rights reports. UNHCR position papers. Congressional Research Service reports. News archives. NGO documentation. Expert declarations.
The best practitioners at large immigration firms and organizations like Human Rights First produce country conditions briefs that run 40 to 80 pages with hundreds of citations. A solo practitioner or small firm cannot match that output. You pull the State Department report, find a few news articles, cite the UNHCR guidelines, and hope it is enough. Sometimes it is not. The IJ finds the claim insufficiently documented. The BIA affirms. The client is ordered removed.
With Dhakma Core: Dhakma Core arrives pre-loaded with a curated library of State Department human rights reports, UNHCR position papers, CRS reports, and NGO documentation from Human Rights Watch, Amnesty International, and Freedom House. You add your own firm's research archive and prior country conditions briefs. The system indexes everything locally and updates as you add new sources.
When you open a new asylum case, you query: "Document current conditions for LGBTQ individuals in Uganda, including police harassment, societal violence, and government enforcement of the Anti-Homosexuality Act." The system assembles a comprehensive research dossier with specific citations, organized by category, with direct quotes from primary sources. You review, verify the citations, add your legal analysis, and file a country conditions package with the depth of a dedicated research team behind it.
What this unlocks: You win asylum cases you would currently lose because the country conditions brief was not deep enough. A two-attorney firm produces research output that rivals the immigration clinics at law school programs with teams of student researchers. The IJ sees a well-documented case. The trial attorney has less room to argue insufficient evidence of country conditions.
Form preparation and consistency checking
The problem: Immigration practice runs on forms. I-130, I-485, I-765, I-131, I-864, I-589, I-751, I-140, I-129. Each form asks overlapping questions in slightly different ways. An applicant's address history appears on the I-485, the I-589, and the I-765. Their employment history appears on the I-485, the I-765, and the I-864. A single inconsistency between forms triggers an RFE or worse. USCIS adjudicators cross-reference entries. So does ICE trial counsel if the case ends up in removal proceedings.
Your paralegal fills out each form manually. They check and re-check. They still miss things. An RFE costs two to four weeks of processing time and hours of staff time to respond. A noticed inconsistency at an interview can derail an otherwise approvable case.
With Dhakma Core: Feed your intake questionnaire and supporting documents into the system. It pre-fills every applicable form from a single data source. When the same information appears on multiple forms, it populates identically. When it detects a conflict between the intake data and a supporting document, it flags it before you file. Date discrepancies between the birth certificate and the I-130. Address gaps in the I-485 that do not match the lease agreements on file. Employment dates on the I-140 that conflict with the tax returns supporting the I-864.
What this unlocks: RFE rates drop. Interview preparation becomes verification rather than discovery. Your paralegal shifts from data entry to case strategy support. A high-volume family-based practice that files 15 to 20 adjustment packages per month saves 30 to 40 hours of paralegal time on form preparation alone.
Declaration drafting from interview transcripts
The problem: The asylum declaration is the foundation of the I-589. It must be detailed, consistent, and written in the applicant's voice. Most attorneys conduct a one-to-two-hour interview, take notes, and then spend three to five hours drafting the declaration from those notes. The attorney writes in a style that sounds like a lawyer, not like the client. The IJ notices. A declaration that reads like an attorney drafted it from a template carries less weight than one that sounds like the applicant telling their own story.
With Dhakma Core: Record the client interview (with consent). The system transcribes the interview and converts it into a structured, chronological outline that preserves the client's phrasing and sequencing. It identifies gaps that need follow-up questions and flags potential credibility issues where the narrative is internally inconsistent. The attorney reviews, conducts a follow-up interview to fill gaps, and refines. Total time from interview to final declaration: reduced by half or more.
What this unlocks: Declarations that sound like your client, not like your word processor. IJs hear the applicant's voice in the written record before they ever take the stand. And you catch inconsistencies between the declaration and the I-589 before the trial attorney does.
Caseload queries and deadline tracking
The problem: Immigration deadlines are unforgiving. A missed filing window on a NACARA application is permanent. A late I-751 filing after conditional residence expires triggers removal proceedings. An asylum application filed after the one-year deadline requires an exception that most IJs scrutinize heavily. Your case management system tracks deadlines, but it does not answer questions like: "Which of my asylum cases filed in the last 90 days have not yet received interview notices?" or "Which I-751 petitions have biometrics appointments scheduled in the next 30 days?"
With Dhakma Core: Query your full docket in natural language. "Show me every case with an RFE response due this month." "Which clients have EAD renewals expiring in the next 60 days?" "List all asylum cases where the asylum clock has been running for more than 150 days without a scheduled interview." The system searches across your entire case database and returns structured results. No clicking through individual case files. No running reports in three different systems.
What this unlocks: Nothing falls through the cracks. The managing attorney sees the full picture of the practice without relying on a paralegal to compile weekly status reports by hand. When USCIS processing times shift, you adjust your entire docket strategy in minutes, not days.
The security is not theoretical
Dhakma Core includes a physical air-gap switch. This is a hardware disconnect. When engaged, no data enters or leaves the system. No software exploit bridges a physical gap. No firmware vulnerability creates a back door. The machine is electrically disconnected from any external network.
Your client's A-numbers, passport data, persecution narratives, and case strategies exist on one machine in your office. Not on a cloud server in Virginia. Not in a training dataset in Beijing. Not in a breach notification letter you send to 200 asylum applicants explaining that their identities may have been exposed to the governments they fled.
ABA Model Rule 1.6(c) requires reasonable efforts to prevent unauthorized disclosure. For immigration practitioners, the standard should be higher. Your clients trusted you with information that could get them killed. The only reasonable effort is physical control of the hardware.
A solo practitioner with the research output of a 20-attorney firm
This is the core shift. Private AI does not just make your existing workflow faster. It gives you capabilities that were previously reserved for large immigration practices with dedicated country conditions researchers, dedicated paralegals for form preparation, and dedicated case managers tracking hundreds of matters.
A two-person immigration firm with Dhakma Core produces country conditions briefs that match the depth of a university immigration clinic. Files adjustment packages with the consistency checking of a firm that employs a full-time quality control paralegal. Manages a docket of 150 active cases with the oversight that larger firms achieve through layers of staff.
The cost of the system is a fraction of one additional employee. The capability it delivers is equivalent to three.
See it on a real case
Bring a closed asylum case. A completed I-589 with country conditions documentation. A case where you wonder whether a deeper brief might have changed the outcome.
We run it through Dhakma Core on-site. Your hardware. Your office. You watch the system produce a country conditions brief from the source library. You compare it against the brief you filed. You see what the IJ would have seen if your research resources had matched your legal talent.
If the output does not change how you prepare your next asylum case, we take the hardware home.
Your clients trusted you with their lives, not just their cases. The intelligence that protects them should never leave the room.