Skip to main content

Using AI Responsibly in Immigration Legal Services: What ABA Formal Opinion 512 Means for Your Team

By Devin McCaw, FounderPublished 11 min readLeer en español

General information for legal organizations, not legal advice.

ABA Formal Opinion 512, issued July 29, 2024, applies the ethics rules lawyers already follow (competence, confidentiality, communication, candor toward courts, supervision, and fees) to generative AI tools, "whether general purpose or designed specifically for the practice of law." For an immigration team, that means checking every output before relying on it and keeping client information out of tools that learn from it unless the client gives informed consent. It also means a person reviews anything a tool produces before it reaches a client or is filed with USCIS or the immigration court.

In ordinary work, a paralegal pastes interview notes into a free chatbot to draft an asylum declaration, or a volunteer asks one whether an old shoplifting arrest matters for naturalization. Most of that client information came in at intake, and the intake guide covers how to protect what clients tell you there. Page numbers below refer to the opinion itself.

What is Opinion 512, and does it bind your organization?

Opinion 512 comes from the ABA's ethics committee and interprets the Model Rules of Professional Conduct, which the ABA calls "models for the ethics rules of most jurisdictions." Lawyers are bound by their own state's rules, and state ethics opinions interpret them. Opinion 512 expects that "state and local bar association ethics committees will likely offer updated guidance" (p. 2). Two examples:

  • California. The State Bar replaced its 2023 practical guidance with a 2026 version covering "agentic" systems, which act without a person prompting each step, for example by "autonomously facilitating client intake." Lawyers must not let such a system "make substantive legal determinations, communicate legal advice, prepare and file pleadings, or otherwise act in a representative capacity without meaningful lawyer supervision and review."
  • Florida. Ethics Opinion 24-1, dated January 19, 2024, notes that "Advisory ethics opinions are not binding," then says lawyers "may use generative artificial intelligence" if they protect confidentiality, stay accurate and competent, bill properly, and follow the advertising rules.

Some courts also require lawyers to disclose AI use (p. 10), so check the local rules wherever you file.

Accredited representatives answer to federal regulations. 8 CFR 292.3 subjects any "practitioner" before DHS, including "any representative as defined in 8 CFR 1.2," to discipline on the grounds in 8 CFR 1003.102, which also apply in immigration court. One ground is failing "to provide competent representation to a client." The organization's AI policy still belongs to its supervising attorney, and Model Rule 1.0(c) counts "lawyers employed in a legal services organization" as a firm.

Competence (Rule 1.1)

Under Rule 1.1, lawyers "must have a reasonable understanding of the capabilities and limitations of the specific GAI technology that the lawyer might use" (pp. 2 to 3), and keeping it current "is not a static undertaking" (p. 3). One risk the opinion names is "hallucinations," or "ostensibly plausible responses that have no basis in fact or reality" (p. 3). Relying on output "without an appropriate degree of independent verification or review of its output" could violate the duty of competence (pp. 3 to 4).

How much checking is enough "will necessarily depend on the GAI tool and the specific task that it performs" (p. 4); the opinion's example is testing a summarizing tool on a small set of contracts first. Before trusting a tool with RFEs, run it on 10 closed RFEs where you know what USCIS asked for.

Declarations need a line-by-line check every time. A tool drafting an I-589 declaration from interview notes will clean up the English, and it can also fill a gap with a plausible date or town the client never mentioned. Under 8 U.S.C. 1158(b)(1)(B)(iii), a credibility determination may rest on "the consistency between the applicant's or witness's written and oral statements," even when an inconsistency does not go "to the heart of the applicant's claim." Compare the draft with the notes, then read it back to the client in the client's language before anyone signs. However much you review, "the lawyer is fully responsible for the work on behalf of the client" (p. 4).

Confidentiality (Rule 1.6)

Before anyone types information about a client's case into a tool, lawyers "must evaluate the risks that the information will be disclosed to or accessed by others outside the firm," and by people inside it who won't protect it properly (p. 6). One factor is "the sensitivity of the information" (p. 6), and immigration files hold facts like a prior removal order or the details of an asylum claim.

The firmest line concerns "self-learning" tools, which can absorb what one user enters and reveal it later to others. For those, "a client's informed consent is required prior to inputting information relating to the representation," and "merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient" (p. 7). Brainstorming without client information needs no consent (p. 7). Deleting names doesn't always make facts anonymous; see the knowledge article's section on de-identification.

Lawyers "should read and understand the Terms of Use, privacy policy, and related contractual terms and policies of any GAI tool they use," or consult someone who has (p. 7). Opinion 512 also applies Rule 5.3's vendor due diligence to "GAI providers and tools" (p. 11). Ask any vendor these questions, and keep the answers in writing:

  • Do you use what we enter to train or improve your models?
  • Who at your company, or at your subcontractors, can see it?
  • Can we enforce your confidentiality promise, and will you tell us about a breach or a legal demand for client information?
  • Do you keep our information after we leave, or claim any rights to it?

Communication (Rule 1.4)

Opinion 512 says the facts of each case decide whether Rule 1.4 requires telling a client about AI use, and "client disclosure may be unnecessary" in some circumstances (p. 8). You do have to tell the client when:

  • The client asks whether AI was used, or the engagement agreement requires disclosure.
  • You plan to enter information about the representation into a tool that requires the client's informed consent.
  • AI use is relevant to the basis or reasonableness of your fee.
  • The tool's output "will influence a significant decision in the representation" (p. 8).

The last one comes up in removal defense. If a tool's estimate of a client's chances feeds the decision whether to seek cancellation of removal, the client would "reasonably want to know" whether the lawyer "is exercising independent judgment or, in the alternative, is deferring to the output of a GAI tool" (p. 9). The engagement agreement is "a logical place" for disclosures (p. 9), written in plain words and in the client's language.

Candor and meritorious claims (Rules 3.1, 3.3, and 8.4(c))

Rule 3.1 requires a non-frivolous basis in law and fact for every issue a lawyer raises. Rule 3.3 bars knowingly making "a false statement of fact or law to a tribunal," and Rule 8.4(c) prohibits "conduct involving dishonesty, fraud, deceit or misrepresentation." Opinion 512 adds that "even an unintentional misstatement to a court can involve a misrepresentation under Rule 8.4(c)" and lists the problems seen so far: "citations to nonexistent opinions, inaccurate analysis of authority, and use of misleading arguments" (p. 10).

In Mata v. Avianca, Inc., No. 22-cv-1461 (S.D.N.Y. June 22, 2023), the plaintiff's lawyers filed an affirmation citing six decisions ChatGPT had generated, none of which existed. Judge Castel wrote that "there is nothing inherently improper about using a reliable artificial intelligence tool for assistance," but that "existing rules impose a gatekeeping role on attorneys to ensure the accuracy of their filings." The sanctions turned on what came next: when the other side and then the court questioned the cases, the lawyers "doubled down and did not begin to dribble out the truth until May 25." The court found bad faith, imposed a $5,000 penalty on the lawyers and their firm jointly and severally, and ordered letters to their client and to each judge falsely named as the author of a fake opinion.

EOIR's Policy Memorandum 25-40, effective August 8, 2025, says "EOIR has neither a blanket prohibition on the use of generative AI in its proceedings nor a mandatory disclosure requirement regarding its use," though individual adjudicators or courts may adopt standing orders. It warns that practitioners submitting "hallucinated or erroneous AI-generated content in filings before EOIR likely violate professional conduct obligations," and cites grounds in 8 CFR 1003.102, including (c), "knowingly or with reckless disregard offering false evidence," and (u), repeated EOIR filings that show "little or no attention to the specific factual or legal issues applicable to a client's case." The regulation also covers DHS filings, which the memo omits, so a pattern of generic AI-drafted briefs or RFE responses could fit (u). Before anything goes to USCIS or the court, someone should open every cited decision, regulation, and Policy Manual chapter and confirm it says what the filing claims.

Supervision (Rules 5.1 and 5.3)

Under Rules 5.1 and 5.3, "managerial lawyers must establish clear policies regarding the law firm's permissible use of GAI," and supervisory lawyers must make reasonable efforts, including training, to see that lawyers and nonlawyers follow them (p. 10). A footnote suggests marking "all materials produced by GAI tools" as such in client and firm files "so future users understand potential fallibility of the work" (p. 10, note 52).

Volunteers often work from their own phones and laptops, so the policy has to reach them before a first shift. The knowledge article covers what Rules 5.1 and 5.3 ask of supervisors, and the AI policy belongs in that training.

Fees (Rule 1.5)

This part matters mostly to firms, and fee rules vary by state. Under Rule 1.5, lawyers who bill hourly "must bill for their actual time," which includes time spent entering information into a drafting tool and reviewing the draft (p. 12). If a tool makes the work much faster, "it may be unreasonable under Rule 1.5 for the lawyer to charge the same flat fee when using the GAI tool as when not using it" (p. 12), so a firm with a flat fee for I-130 packets should check that the fee still matches the work. A lawyer also "may not charge a client to learn about how to use a GAI tool or service that the lawyer will regularly use for clients" (p. 14).

Why do client-facing AI tools need extra care?

People asking a client-facing chatbot may not know the difference between general information and advice about their own case. Someone types "Do I qualify for citizenship?" A tool that counts years since the green card may say yes and miss a 7-month trip abroad: under 8 CFR 316.5(c)(1)(i), absences "for continuous periods of between six (6) months and one (1) year" disrupt continuous residence "unless the applicant can establish otherwise to the satisfaction of the Service." Applying the law to this person's facts is legal advice, and lawyers "may not leave it to GAI tools alone to offer legal advice to clients" (p. 4). The case-status article draws the same line for status updates.

Florida's Opinion 24-1 warns that an intake chatbot can create a prospective client relationship, or even a lawyer-client relationship, "without the lawyer's knowledge," and tells lawyers to be wary of "an overly welcoming generative AI chatbot that may provide legal advice, fail to immediately identify itself as a chatbot, or fail to include clear and reasonably understandable disclaimers limiting the lawyer's obligations." It is written for Florida lawyers, and its reasoning is a sound starting point for any client-facing tool:

  • The tool says it is a computer program, in the client's language, before its first answer.
  • General answers cite their source.
  • A question that turns on the person's own facts goes to a person, and an attorney or accredited representative approves any individualized reply before the client sees it.
  • Every question and reply is saved to the client's file.

An AI-use checklist for immigration teams

  • The supervising attorney approves a written list of AI tools and the tasks each may be used for.
  • Client names, A-Numbers, receipt numbers, and case facts stay out of consumer chatbots, and go into self-learning tools only with the client's informed consent.
  • Every citation, quote, and date in AI-assisted work is checked against its source before anything is filed or sent.
  • AI-drafted documents are marked as such in the client file.
  • A person approves any client-facing reply that applies the law to the client's facts.
  • A log records which tool was used on which matter.
  • Staff and volunteers are trained before a first shift and every year after.
  • The engagement agreement explains AI use in plain words, in the client's language.
  • Vendor answers on training, retention, and breach notice are on file.
  • The policy is reviewed on a set date, such as every 6 months, against new state bar guidance and court or EOIR standing orders.

Where Parasource fits

Parasource provides legal information only, never legal advice, and its answers cite their sources so a person can check them. Its internal caseworker assistant gives staff citation-grounded answers that are treated as staff work product and checked by a person. Client data is encrypted at rest, and no organization's knowledge is shared with another.

An organization can also opt in to a client responder, where a lawyer or accredited representative approves each reply. It can choose, category by category, to let generic informational replies send automatically; individualized replies never auto-send. None of this replaces your own AI policy or your supervising attorney's review.

Sources

See how Parasource handles intake

Parasource gives immigration organizations multilingual intake, a private knowledge base built from their own work, and an opt-in client responder that holds every individualized reply for your legal team's approval. Book a walkthrough to see it with your own forms.

Book a demo

Keep reading