Your Firm Needs an AI Policy: A Michigan Supervision Guide
Most of the coverage about artificial intelligence and lawyers focuses on the person who signed the bad brief. That framing misses half the exposure. When an associate, a paralegal, or a contract attorney files fabricated citations, the partner who runs the firm can be on the hook too, not because they touched the document, but because they never put a policy or a training program in place. This is a plain-English guide for Michigan managing and supervising lawyers on what Rules 5.1 and 5.3 now require, why a written AI policy has moved from nice-to-have to a professional duty, and what the policy should actually say.
Supervision Liability Is Not Theoretical Anymore
A federal case out of Wyoming made the point concrete. After a firm filed motions that cited eight cases that did not exist, the court did not stop at the lawyer who drafted the filing. It fined the drafting attorney $3,000 and revoked his admission to appear in the case, then separately fined two other attorneys who had signed the filing, Michael Morgan and local counsel Taly Goody, $1,000 each. See the reporting at Spellbook’s summary of the sanctions order. The mechanism there was the signature itself. Every lawyer who signs a filing certifies it, so the court reached the co-signers, not only the drafter. For anyone who runs a firm, that lesson carries straight into the professional-conduct rules that govern oversight.
Michigan practitioners should not read that as a distant, out-of-state problem. In March 2026 the Sixth Circuit, the federal appellate court that covers Michigan, fined two attorneys $15,000 each over a brief that contained more than two dozen fabricated citations of the kind associated with unverified AI output. See the account at LawSites. Sanctions of that size are no longer the exception, and firms that put nothing in writing are the ones most exposed when a filing goes wrong.
What Rules 5.1 and 5.3 Actually Say
The obligations here predate generative AI by decades, which is exactly why lawyers keep getting caught off guard. Michigan Rule of Professional Conduct 5.1 places a structural duty on partners and lawyers with managerial authority to make reasonable efforts to ensure the firm has measures in place that give reasonable assurance every lawyer follows the rules. It is not enough to be personally careful. A managing lawyer has to build a system.
Rule 5.3 extends the same duty to nonlawyer assistance. The American Bar Association changed its model rule’s title in 2012 from “Nonlawyer Assistants” to “Nonlawyer Assistance,” and ABA guidance and several state bars now read that broader word choice to reach technological tools, not only human staff. Michigan’s own rule still carries the older title, so the “covers AI tools” reading is best treated as the prevailing interpretation rather than settled black-letter text. The practical takeaway holds either way: an AI tool that drafts a memo functions like a junior clerk who drafts a memo, and the supervising lawyer owes the output the same review.
Two more Michigan rules give this teeth. MRPC 1.1 requires competent representation, and a 2019 comment adopted by the Michigan Supreme Court makes clear that competence includes keeping abreast of changes in the law and its practice, including the use of existing and developing technology, to the extent reasonably necessary for the representation. The State Bar of Michigan has since published an Artificial Intelligence FAQ and a June 2025 Age of AI report confirming that competence now reaches generative tools. A firm cannot supervise a technology its lawyers do not understand, so the competence duty and the supervision duty run together.
ABA Opinion 512 Turns the Duty Into a Checklist
On July 29, 2024, the ABA issued Formal Opinion 512, its first ethics guidance on generative AI. Michigan has not issued a formal opinion of its own, so 512 is the clearest available roadmap, and it maps onto Michigan’s parallel rules cleanly. On supervision, the opinion is blunt. It states that managerial lawyers must establish clear policies about the firm’s permissible use of generative AI, and that lawyers with supervisory authority must make sure nonlawyers, including outside contractors, are adequately trained in the ethical and practical uses of the technology. See the ABA’s own explanation in Business Law Today.
Opinion 512 even describes what that training should cover: the basics of how the tools work, their capabilities and limitations, the ethical issues in use, and best practices for secure handling of client data. That is close to a syllabus. A firm that adopts a short written policy and runs one training session covering those four areas has done most of what the opinion asks. A firm that does none of it is relying on luck.
| Rule | Who it binds | What it requires for AI |
|---|---|---|
| MRPC 5.1 | Partners and managing lawyers | Put firmwide measures in place, including a written AI policy, that give reasonable assurance the rules are followed |
| MRPC 5.3 | Any lawyer supervising nonlawyers or tools | Train and oversee paralegals, staff, and AI tools; treat AI output like a junior’s work product |
| MRPC 1.1 | Every lawyer | Understand what the tools do well and where they fail; verify output before it is used |
| MRPC 1.6 | Every lawyer | Obtain informed client consent before entering confidential facts into a tool that may store or train on them |
| MRPC 5.3 (delegation) | Supervising lawyer | Cannot delegate the verification duty to the tool or the junior; the review is the supervisor’s own |
What a Michigan Firm AI Policy Should Cover
A policy does not need to be long to satisfy the rules. It needs to be clear, written down, and actually followed. The firms that get into trouble usually have no document at all, which leaves every lawyer to invent private habits and leaves the managing partner with nothing to point to. A workable policy addresses six areas.
- Approved tools. Name which systems the firm permits and which it forbids. A closed, firm-controlled platform is different from a public chatbot that trains on whatever is typed into it, and the policy should say so plainly.
- Confidentiality boundaries. State that client identifying facts do not go into public tools without informed consent, and give staff a safe default: strip identifiers or use the approved closed system.
- Mandatory verification. Require that every citation and factual claim in any AI-assisted work be independently confirmed in a primary source before it reaches a signature line. Name who is responsible for that check on each matter.
- Disclosure and billing. Address whether and how the firm discloses AI use, and make clear the firm bills for work performed, not for time a tool saved, consistent with Opinion 512.
- Training cadence. Commit to onboarding training for new hires and a periodic refresher, covering the four areas Opinion 512 identifies.
- Accountability. Identify who owns the policy, how questions get answered, and what happens when the policy is not followed.
None of this requires a technology background. It requires a managing lawyer to decide the firm’s rules once, write them down, and make sure the people doing the work have read them.
Small Firms and Solos Are Not Exempt
It is tempting for a two-lawyer office or a solo with one paralegal to assume the supervision rules are aimed at large firms. They are not. Rule 5.1 binds any lawyer with managerial authority over a practice, and in a small shop that is the owner. Rule 5.3 applies the moment a single nonlawyer, or a single tool, touches client work. The smaller the firm, the more directly the supervision duty lands on one person, because there is no committee to absorb it.
There is a practical upside for small practices. A short policy and a brief training session are easier to implement when there are fewer people to align, and the same document that satisfies the ethics duty also protects the practice from the reputational and malpractice consequences that follow a public sanctions order. Malpractice carriers have started asking about AI use on renewal applications, and a firm that can produce a written policy answers that question from a position of strength.
The Cost of Doing Nothing
A sanctions order is public and permanent. It surfaces in opposing counsel’s research, in a malpractice carrier’s file, and in a prospective client’s internet search for years. When the sanction reaches a supervisor who never built a policy, the finding is not just that a mistake happened. It is that the firm had no system to prevent it. That is a harder thing to explain to a client, an insurer, or a disciplinary panel than a single careless filing. The written policy and the training session are cheap. The absence of them is what becomes expensive.
अक्सर पूछे जाने वाले प्रश्न
Does my law firm need a written AI policy?
In practical terms, yes. ABA Formal Opinion 512 states that managerial lawyers must establish clear policies on the firm’s permissible use of generative AI. Michigan has no formal opinion of its own, but MRPC 5.1 requires managing lawyers to put firmwide measures in place, and a written AI policy is the most direct way to meet that duty and to have something to point to if a filing is questioned.
Are partners responsible for an associate’s AI mistakes?
They can be. MRPC 5.1 and 5.3 make supervising and managing lawyers responsible for building systems and overseeing the people and tools that do the work. In one Wyoming federal case the court fined not only the attorney who drafted the fabricated citations but also two other lawyers who signed the filing. A partner cannot escape a bad filing by pointing at the junior who ran the search.
What should a law firm AI policy include?
At a minimum: which tools are approved and which are forbidden, confidentiality boundaries for client data, a mandatory verification step for every citation and factual claim, the firm’s position on disclosure and billing, a training cadence for new and existing staff, and a named owner for the policy. It does not need to be long. It needs to be written and followed.
Does Michigan require lawyers to supervise AI tools?
Yes, through existing rules. MRPC 5.3 covers nonlawyer assistance, which ABA guidance and multiple state bars read to include technological tools. Michigan attorneys must supervise AI use with the same care they apply to supervising a paralegal, which means verification, oversight, and quality control before the output is used.
Do I have to train my paralegals and staff on AI?
Opinion 512 says supervising lawyers must make sure nonlawyers, including outside contractors, are adequately trained in the ethical and practical use of these tools. The opinion suggests training should cover how the tools work, their limits, the ethical issues, and secure handling of client data. A single documented session covering those areas goes a long way toward meeting the duty.
Is a small firm or solo practice exempt from these rules?
No. MRPC 5.1 binds any lawyer with managerial authority, which in a small office is the owner, and MRPC 5.3 applies the moment one nonlawyer or one tool touches client work. The duty is easier to meet in a small firm because there are fewer people to align, but it is not optional.
Have a Michigan injury case that outgrew a do-it-yourself approach?
Attorney Manny Chahal handles serious personal-injury and no-fault matters across Michigan and accepts referrals and co-counsel arrangements. If a case needs a partner who will do the work and verify it, start a conversation.
1-844-624-2425 पर कॉल करें