Skip to main content

Blog · Legal Ethics

Rule 5.3 and AI: How to Supervise Staff and AI Tools Under Rules 5.1 and 5.3

The supervision rules are where most AI sanctions actually start: a clerk or associate drafts with a chatbot, and the signing lawyer never checks. Here is how Model Rules 5.1 and 5.3 apply to staff who use AI and to the AI vendors themselves, what ABA Formal Opinion 512 and the North Carolina and D.C. opinions say, and a supervision protocol a small firm can run.

Rule 5.3 applies to AI in two ways. It makes lawyers answer for how their paralegals, clerks, and assistants use AI, and, per ABA Formal Opinion 512, its outsourcing analysis extends to the AI vendors themselves. Together with Rule 5.1 for lawyers, it requires clear policies, training, vendor vetting, and a lawyer who reviews the work.

Most AI sanctions so far have been framed as candor or Rule 11 problems. Read the facts, though, and many begin as supervision failures: someone other than the signing lawyer drafted with a chatbot, and nobody with authority checked. This post covers what the supervision rules say, how the ABA and state opinions apply them to AI, and a protocol a small firm can run. It is the supervision analysis; for the policy document itself, use our one-page law firm AI policy.

What Rules 5.1 and 5.3 actually say

The two rules have the same three-part structure. Rule 5.1 covers lawyers; Rule 5.3, titled "Responsibilities Regarding Nonlawyer Assistance," covers anyone else "employed or retained by or associated with a lawyer."

PartWho it bindsDuty under Rule 5.3
(a)Partners and lawyers with comparable managerial authority"make reasonable efforts to ensure that the firm has in effect measures giving reasonable assurance that the person's conduct is compatible with the professional obligations of the lawyer"
(b)A lawyer with direct supervisory authority over the nonlawyer"make reasonable efforts to ensure that the person's conduct is compatible with the professional obligations of the lawyer"
(c)Any lawyer, in two situationsResponsible for the nonlawyer's violation if the lawyer orders or ratifies it, or is a manager or supervisor who knows of it "at a time when its consequences can be avoided or mitigated but fails to take reasonable remedial action"

In a two- or three-lawyer firm, the same person usually holds all three roles. You are the manager who sets the measures, the supervisor who reviews the work, and the lawyer who knew, or should have looked, before the filing went out.

The rule's title matters. The ABA changed it from "Assistants" to "Assistance" in 2012, with the Ethics 20/20 amendments, to reach services from outside the firm as well as employees. That is what lets the rule apply to an AI vendor.

Two kinds of AI supervision

Supervising the people who use AI. A paralegal summarizing depositions with a chatbot, an associate drafting a motion, a law clerk running research: each is conduct the supervising lawyer must make reasonable efforts to keep compatible with the lawyer's obligations. That includes confidentiality (what went into the tool), competence (whether the output was checked), and candor (whether anything false reached a court).

Supervising the tool and its vendor. Opinion 512 says lawyers "have additional supervisory obligations insofar as they rely on others outside the law firm to employ GAI tools," and that Rule 5.3(b) imposes the duty of "reasonable efforts to ensure that" the nonlawyer's conduct conforms. It then lists the diligence from earlier outsourcing opinions: reference checks and vendor credentials, the vendor's security policies, its hiring practices, confidentiality agreements, its conflicts screening, and the availability of a legal forum if the agreement is breached. "These concepts also apply to GAI providers and tools."

North Carolina says the same thing in its own words. 2024 Formal Ethics Opinion 1 holds that Rule 5.3's requirements "apply to nonlawyer assistants within a law firm as well as those outside of a law firm that are engaged to provide assistance in the lawyer's provision of legal services to clients, such as third-party software companies." It places AI "somewhere between" ordinary software and nonlawyer support staff. Our North Carolina AI ethics opinion post covers the rest of it.

What ABA Formal Opinion 512 requires of supervisors

Opinion 512's supervision section (Part E) is short and specific:

  • Policies. "Managerial lawyers must establish clear policies regarding the law firm's permissible use of GAI, and supervisory lawyers must make reasonable efforts to ensure that the firm's lawyers and nonlawyers comply with their professional obligations when using GAI tools."
  • Training. Supervisory obligations include "ensuring that subordinate lawyers and nonlawyers are trained," covering "the basics of GAI technology, the capabilities and limitations of the tools, ethical issues in use of GAI and best practices for secure data handling, privacy, and confidentiality."
  • Vendor terms. Drawing on its cloud-computing opinions, it says lawyers should make sure the tool is configured to preserve confidentiality, investigate its reliability and security, determine whether it retains information after services end or claims rights to it, and understand that the vendor's servers can fail or be attacked.
  • Labeling. A footnote offers one training suggestion: that all materials produced by AI tools "be marked as such when stored in any client or firm file so future users understand potential fallibility of the work."

What D.C. Opinion 388 adds

The D.C. Bar's Opinion 388 (April 2024) goes further on what supervisors may permit. It says that if a review of a tool's privacy policy shows the tool will not keep client information from its owner and outsiders, "Rules 5.1 and 5.3 likely prohibit the supervising lawyers from permitting supervised personnel to use the GAI in the client matter." It suggests requiring staff to verify AI output, and says that once a firm has vetted a tool for particular purposes, "it could require its lawyers to use that GAI for those purposes rather than some competing GAI not yet approved within the firm." Our D.C. Opinion 388 explainer covers the full opinion; other states are on our state bar AI guidance map.

What a supervision failure looks like

The clearest case so far is Utah's. In Garner v. Kadince, Inc., 2025 UT App 80, a petition cited fabricated authority obtained from ChatGPT. Counsel explained that it "had been prepared by an unlicensed law clerk, employed by their law firm," that the signing lawyer "did not independently check the accuracy of the Petition before signing and filing it," and that the firm "did not have any policy in place addressing the use of AI." The court held that counsel "fell short of their gatekeeping responsibilities" and ordered them to pay the other side's fees, refund their client, and donate $1,000 to a legal-aid organization.

Every element of a Rule 5.3 failure is in those facts: no measures in place under 5.3(a), no review under 5.3(b), and a signing lawyer who adopted the work as his own. The court's remedy ran against the lawyers, not the clerk. Our guide to Utah AI ethics for lawyers covers the case in its local context.

A supervision protocol for a small firm

A policy says what is allowed. A protocol says who checks what, and when. This one maps each step to the rule it satisfies.

1. Approve tools before anyone uses them (5.3(a)). Run Opinion 512's vendor questions on each tool: security, confidentiality terms, retention after termination, whether the vendor claims rights to inputs, and whether it trains on your data. Record the answers. Personal accounts on consumer chatbots stay off client work.

2. Assign every AI-assisted task to a named supervising lawyer (5.3(b), 5.1(b)). The paralegal or associate using the tool knows who reviews the output. "The firm" is not a reviewer.

3. Set review by task type. Opinion 512 says the level of review depends on the tool and the task. A practical tiering:

TaskMinimum review before use
Brainstorming, outlines, internal notesSupervisor skims for confidentiality and obvious error
Summaries of documents or depositionsSpot-check against the source; full check if it will be relied on
Client advice or correspondenceSupervising lawyer reads it in full and owns the judgment
Anything filed with a courtEvery citation, quotation, and pin cite verified by the signing lawyer

4. Label AI work product in the file. Follow Opinion 512's footnote: mark AI-produced materials so the next person knows to verify them.

5. Train once, then refresh. One session covering how the approved tools produce their answers, what client information may go in, and the review tiers. Repeat when tools change. Our guide for paralegals doing legal research is a reasonable pre-read for staff.

6. Spot-check. Supervision is ongoing. Once a month, pull one AI-assisted work product at random and recheck it end to end.

7. Act when something goes wrong (5.3(c)(2)). If you learn of a problem "at a time when its consequences can be avoided or mitigated," the rule requires reasonable remedial action. For a filing, that means correcting the record promptly under Rule 3.3, then finding how the error got through.

Step 3's last row is the one that produces sanctions, and it is the easiest to support with software. Our brief cite-checking guide gives the manual method. CaseRead's research cites only sources it actually retrieved from public law and, when you want, your own matter files, and the citation check runs on drafts written anywhere.

Frequently asked questions

Does Rule 5.3 apply to AI?

Yes, in two ways. It applies to the people in your firm who use AI, such as paralegals, clerks, and assistants. ABA Formal Opinion 512 also applies Rule 5.3's outsourcing analysis to AI providers and tools, so the vendor diligence lawyers already do for outside services, including security, confidentiality agreements, and retention terms, applies to the AI vendor.

What do Rules 5.1 and 5.3 require of a small-firm lawyer using AI?

Lawyers with managerial authority must have measures in place giving reasonable assurance that lawyers and staff comply with the ethics rules, and supervising lawyers must make reasonable efforts to ensure the people they supervise do. ABA Opinion 512 reads that as clear policies on permissible AI use, training, and vetting of outside AI providers.

Is the supervising lawyer responsible if a paralegal files AI-generated fake citations?

Often, yes. Under Rule 5.3(c), a lawyer is responsible for a nonlawyer's conduct if the lawyer ordered or ratified it, or if a partner or supervisor knew of it in time to avoid or mitigate the consequences and failed to act. Separately, the lawyer who signs a filing certifies its contents, as the Utah Court of Appeals held in Garner v. Kadince.

Should a firm ban staff from using unapproved AI tools?

D.C. Ethics Opinion 388 suggests that once a firm has vetted a tool for a purpose, it could require staff to use that tool rather than an unapproved one, and that Rules 5.1 and 5.3 likely prohibit letting staff use a tool whose privacy policy would expose client confidences. An approved-tools list is the practical way to do that.

Make the last check automatic

The supervision rules ask for reasonable efforts, and the cheapest reasonable effort is checking every citation before a filing leaves the office. Have the signing lawyer paste the draft or upload the file into CaseRead's free Hallucination Shield: it checks up to 50 citations per run against official statute and case-law sources, with no signup. For firm-wide research with verified citations, see CaseRead's plans.