Is AI Safe for Immigration Law Firms?

Every immigration attorney I talk to asks some version of the same question before they adopt AI. Is this safe? It is the right question, and the honest answer is that safety is not a property of "AI." It is a property of the specific tool you choose and how you use it.

I am an immigration attorney, and our clients hand us the most sensitive material imaginable. Asylum trauma narratives, medical and financial records, candid immigration histories. The facts that build a persuasive filing are the same facts that do real damage if they leak or become discoverable.

Why "Is AI Safe?" Is Really a Question About the Tool, Not the Technology

The clearest guidance we have came from a courtroom. In United States v. Heppner, a federal court held that a defendant's exchanges with a public, consumer AI chatbot were protected by neither attorney-client privilege nor the work-product doctrine. The problem was not that AI was involved. The problem was the tool class and who controlled it. A public tool, used without the lawyer's direction, under terms that let the platform retain and disclose inputs, never had the conditions privilege depends on. We break the ruling down in our analysis of AI and attorney-client privilege after Heppner.

That reframes the safety question entirely. You are not asking "is AI safe." You are asking "is this tool attorney-directed, built for legal work, and run by a vendor whose terms protect my data." Those are answerable questions, and a good vendor will answer them in writing.

The Three Things That Actually Determine Whether Client Data Stays Protected

Strip away the marketing and three factors decide whether a tool keeps client information protected.

1. Who directs the tool. Privilege depends on the work being attorney-directed. A tool you control as part of your practice can function as an extension of your work, the way a paralegal or retained expert does. A client pasting facts into a free chatbot on their own is the fact pattern that lost in Heppner.
2. What the vendor's terms permit. If the terms let the platform train on your inputs or disclose them to third parties, you have a confidentiality problem no feature can fix. Consumer tiers often allow exactly that.
3. How the data is secured. Encryption in transit and at rest, sensible retention, and the ability to delete your data are the table stakes that keep information from being exposed in the ordinary course.

The Vendor Security Checklist: Questions to Ask Before You Buy

Send these to any AI vendor before client facts touch the tool. A vendor built for legal work will answer them quickly and put the answers in your contract. Hesitation is itself an answer.

• Do you train your models on our inputs or outputs?
• Do you share, sell, or disclose our data to third parties?
• Is our data encrypted in transit and at rest?
• How long do you retain our data, and can we delete it on demand?
• Where is our data stored, and who are your sub-processors?
• Will you sign a confidentiality agreement or data processing agreement?
• Was this tool built for legal or professional use, or is it a consumer product?

If you want the wider context on choosing among tool types, our guide to the best AI tools for immigration lawyers maps the categories, and our piece on the ethical challenges of AI adoption covers the duty-of-competence framing.

Client-Side Risk: The Part Most Firms Miss

You can vet your own tools perfectly and still have a leak, because the Heppner fact pattern was a client acting alone. Immigration clients face language barriers and a strong self-help instinct, and many turn to free chatbots between consultations to "understand" their case, pasting in the exact facts you are trying to protect.

Two moves close that gap. First, tell clients in writing not to enter sensitive case facts into public AI tools, educate them on the Heppner ruling, and add an AI-use clause to your retainer. Our AI use policy template for immigration firms gives you language to adapt. Second, make sure your own staff are using firm-approved tools rather than personal accounts, which is the shadow-AI problem that same template is built to solve.

A Quick Self-Audit for Your Firm

• Every AI tool the firm uses is on an approved list, signed in with firm accounts.
• Each vendor says, in writing, that they do not train on or disclose your data.
• Your retainer includes an AI-use clause and you have told clients not to use public AI on their case.
• A human verifies every AI-assisted work product, and an attorney signs off before anything is filed.

Frequently Asked Questions

Is it ethical to use AI for immigration cases?

Yes, within the duty of competence. You must understand what the tool does, verify its output, protect client confidentiality, and keep decision authority. The ABA's guidance on AI and the Florida Bar's AI ethics opinion are aligned on these principles.

Does using a secure AI tool guarantee attorney-client privilege?

No. The right tool class preserves the conditions privilege depends on, but your jurisdiction's rules and your own professional judgment still govern. Keep the workflow attorney-directed and document client consent where the rules require it.

What is the safest type of AI tool for immigration work?

A purpose-built, attorney-directed tool from a vendor that contractually will not train on or disclose your data and encrypts it.

Can client data ever be truly confidential in an AI tool?

It can be protected to the standard the rules require when the vendor's terms forbid training and disclosure, the data is encrypted, and you control the workflow. That is a different posture from a consumer tool whose terms allow it to learn from and share what you type.

Bottom Line

AI is safe for immigration work when you treat the choice of tool as a confidentiality decision, not a convenience one. Ask the vendor the questions above, get the answers in your contract, keep clients off public chatbots, and keep an attorney in control. Drafty AI was built for that standard. It is purpose-built for immigration, attorney-directed, does not train on your data, and encrypts it in transit and at rest. Start a free trial and put it through the checklist yourself, or see pricing.