Lesson objective
Explain the structural features of a partnership that spread responsibility for technology decisions until nobody holds it, and locate where in your own firm an AI decision currently comes to rest.
Plain-language explanation
Why this matters now
Level 3 ended with an operation: a workflow that had an owner, a written method, a measure that could get worse, and an escalation path somebody had tested. That is the largest unit one practice group leader can hold in their head.
This level begins one level up, and the question changes shape. It is no longer "is this workflow sound?" It is the question a client's general counsel asks in a questionnaire, an insurer asks at renewal, and a disciplinary authority asks after something has gone wrong: who in this firm is answerable for the use of artificial intelligence, and how would you show me?
Most firms cannot answer it, and the reason is not negligence. It is structure. A partnership is an arrangement of co-equal owners who have agreed to share profit and overheads while retaining autonomy over their own practice. That arrangement is very good at distributing capital, work and risk. It is very bad at concentrating responsibility for something that crosses every practice group at once.
Law / rule: Model Rules 5.1 and 5.3 place responsibility on partners and lawyers with comparable managerial authority to make reasonable efforts to ensure the firm has measures in effect giving reasonable assurance that lawyers and non-lawyer assistants conform to the rules of professional conduct. These are model provisions. Every jurisdiction adopts its own version, with its own numbering and wording, and some adopt none of it; read the rules of every jurisdiction in which your firm practises, and say so to your partners every time you cite a number.
Professional guidance: ABA Formal Opinion 512 (29 July 2024) treats firm policies on permissible generative-AI use as part of how managerial and supervisory lawyers discharge those duties, alongside competence, confidentiality and informed consent, candour, and billing only time actually expended. A model rule tells you that somebody must be answerable. It does not tell you who, in your firm, that is.
A question to sit with
Name the single person in your firm who would have to answer, in their own name, for a decision to allow an artificial-intelligence tool to touch client material. If you cannot name one, that is not an administrative gap. It is the finding.
What goes wrong
Equality is mistaken for accountability. In a partnership every partner is an owner, so every partner feels entitled to an opinion and none feels obliged to decide. A question that belongs to everybody belongs to nobody, and the technology question is the purest example because no single practice group's revenue depends on resolving it this quarter.
The decision is delegated downwards into people who cannot make it. Technology questions travel naturally to the chief information officer, the knowledge manager or the legal-operations lead — capable people who are not, in most firms, licensed to make a judgement about a professional duty and who have no authority to tell a partner no. The firm then has a decision-maker with the expertise and no authority, and an authority with neither the expertise nor the time.
Committees absorb accountability without holding it. A technology committee meets, discusses, minutes the discussion and approves a direction. Nobody on it is the person who is answerable, because a committee has no professional licence and cannot be reprimanded. Accountability that has been handed to a body that cannot hold it has been destroyed, politely.
Practice-group autonomy fragments the position. Three groups reach three different answers in good faith. The firm now has no position, and at the same time believes it does, because each group believes its own answer is the firm's.
The originating partner outranks the process. The partner who brought in the client is the person who decides how that client's work is done. Where a firm's governance depends on persuading a rainmaker, the governance is advisory and should be described that way.
The practical answer
Start by separating three things that firms routinely collapse into one word.
| Word | What it means here | Who can hold it |
|---|---|---|
| Responsibility | Doing the work correctly | Anyone doing the work |
| Accountability | Being answerable for the outcome, by name, to someone with standing to ask | A named person, almost always licensed |
| Authority | Being able to say yes and no, and have it stick | A named person or role, defined in writing |
The failure in most firms is that responsibility is everywhere, accountability is nowhere, and authority sits with whoever is most senior in the room at the time. Governance is the work of putting accountability and authority in the same named place and writing it down.
Best practice: apply four tests to any candidate for accountability in your firm.
- Named. A person, not a role, not a committee, not "the firm". Roles are how you describe it in the charter; a name is how you operate it.
- Licensed where the judgement is professional. A decision that turns on confidentiality, competence, candour or supervision must rest with someone admitted to practise. Technical staff advise; they do not absorb a professional duty.
- Authorised. The named person can stop a use, refuse a tool and overrule a partner within a defined scope, and the partnership has agreed in advance that they can.
- Resourced and reachable. Accountability without time, budget or a deputy is a title. If the answer to "who covers when they are away?" is silence, the design has a single point of failure.
The outsourcing analogy makes the duty legible to partners who resist the technology framing. Professional guidance: ABA Formal Opinion 08-451 (5 August 2008) addressed a lawyer's obligations when outsourcing legal and non-legal support services, and concluded that a lawyer who delegates work to an outside provider remains responsible for it and must supervise. Formal Opinion 512 applies that same logic to generative AI. This is useful in a partners' meeting because it moves the question off novelty. The firm already knows how it supervises an outside provider — it has done it for decades with contract reviewers, e-discovery vendors and translators. AI is another provider whose work reaches a client over a lawyer's name.
Use a management-system vocabulary so the discussion has somewhere to land. Best practice: the four functions of the NIST AI Risk Management Framework 1.0 (26 January 2023) — Govern, Map, Measure and Manage — give a partnership an agenda it can work through without anyone having to invent one. Govern is this module and the next: who is answerable, and under what instrument. Note carefully that the framework is voluntary; adopting it neither creates a professional duty nor discharges one. The same caution applies to ISO/IEC 42001:2023 (18 December 2023), the certifiable AI management-system standard. A firm can run a genuine management system without either, and a firm can adopt both and still be unable to say who approved anything.
Decide where the decision comes to rest, and test it with three live questions. Take these to the next partners' meeting exactly as written.
| Question | If the honest answer is a name | If the honest answer is a shrug |
|---|---|---|
| Who approves a new AI tool touching client material? | Record it; you have a decision point | You have shadow adoption already, and no visibility into it |
| Who can tell a partner that a use is not permitted? | Confirm the partnership has agreed this in writing | Your governance is advisory; describe it that way |
| Who answers a client questionnaire about the firm's AI use? | Check they can produce evidence, not intentions | The first questionnaire will be answered by whoever is free |
Remember this
- A partnership distributes ownership and autonomy by design; that same design distributes responsibility for firm-wide technology until nobody holds it.
- Responsibility, accountability and authority are three different things, and governance means putting the last two in the same named place.
- Law / rule: Model Rules 5.1 and 5.3 say somebody must be answerable but never who; your jurisdiction's adopted version is the one that binds you.
- A committee, a role with no name, and an unlicensed technologist are three ways of appearing to hold accountability without holding it.
- NIST AI RMF 1.0 and ISO/IEC 42001 are voluntary; they organise the conversation and neither creates nor discharges a professional duty.
Legal example
Harrowgate Veldt LLP is a fictional 120-lawyer firm with four practice groups. Its invented managing partner, Alastair Nkemelu, is asked by a longstanding client to complete an outside counsel questionnaire with a new section on artificial intelligence. Question three asks who at the firm is accountable for its AI programme. Alastair does not know, so he asks. The chief information officer, a fictional Marguerite Sowande, says she is accountable for the tools and their security, but not for whether a lawyer should be using one on a matter; she has raised this twice and been told it is a professional question. The chair of the technology committee, an invented corporate partner named Ivo Brandsma, says the committee approved the firm's AI direction eighteen months ago and forwards the minutes; the minutes record a discussion and a decision to "encourage responsible adoption", and name nobody. The head of litigation says his group has its own rule, written by a senior associate, which permits a drafting assistant but no research tool. The head of the regulatory group did not know there was a firm position and has been running a client-facing summarisation workflow since the spring. The head of employment banned everything a year ago after an article alarmed her, and her group now uses a consumer product on personal accounts because the ban came with no alternative. Alastair puts the five answers side by side and sees the shape of the problem: the firm has five positions and believes it has one. He also sees that the questionnaire cannot honestly be answered. He drafts a reply that says the firm is in the process of naming a single accountable partner and will confirm within thirty days, and then does the harder thing, which is to take the naming to the partnership rather than announce it. The partners' meeting takes ninety minutes and turns, not on the technology, but on whether a named partner may overrule an originating partner about how that partner's own client work is done. The answer the partnership reaches — yes, within a defined scope, with an appeal to the executive committee — is the first piece of real governance the firm has.
Responsible-use warning
Never let a committee be the answer to "who is accountable". A committee has no licence, cannot be reprimanded, cannot be sued in its own name and cannot be reached at seven on a Friday evening. Handing accountability to one feels like resolution and is in fact its opposite: the question stops being asked, the minutes look like governance, and there is nobody whose own name is on the outcome. A committee is a perfectly good place to deliberate, advise, and hold a decision-maker to account. Name the person the committee advises, and write the name down.
Human-review requirement
A licensed attorney owns every output that reaches a client or a court, and no governance structure moves that ownership anywhere. At firm level the reviewing partner additionally owns two judgements that cannot be delegated to technical staff: whether a proposed use is consistent with the duties of competence, confidentiality, candour and supervision as adopted in the relevant jurisdiction, and whether the firm can actually supervise the use at the volume proposed. Where the honest answer to the second is no, the attorney narrows the scope or declines the use rather than approving it and hoping supervision keeps up. Technical staff advise on how a tool behaves; they never absorb a professional duty.
Practical exercise
Map where accountability for AI currently rests in your own firm. Use only non-confidential material, and invent a firm if you would rather not use your own.
- Write the names of every person or body in your firm who has ever said anything approving or restricting AI use: individuals, committees, practice-group leaders, IT, risk, anyone.
- Beside each, mark whether they hold responsibility (they do work), accountability (they are answerable by name) or authority (they can say no and have it stick). Most entries will have exactly one mark. Note every entry with authority but no accountability, and every entry with accountability but no authority.
- Run the corridor test. Ask five people in different parts of the firm who decides whether an AI tool may be used on client work, and record the five answers verbatim without correcting anyone.
- Write the firm's current position on AI in one sentence, as an outsider would have to infer it from behaviour rather than from documents. If you cannot write one sentence, write the three or four positions that actually exist and which group holds each.
- Identify the highest-consequence use currently running anywhere in the firm, and write who approved it, when, and on what evidence. If nobody did, write that.
- Draft the four-test profile for a single accountable person: named, licensed, authorised, resourced. Write what each of the four would require your partnership to agree to, and which of the four would be hardest to get.
- Write the sentence you would need the partnership to vote on. One sentence, naming a role and a scope of authority. Bring it to module 2, where it becomes the opening clause of the governance instrument.
Knowledge check
-
Why does a partnership structure diffuse responsibility for AI rather than concentrate it?
Answer guidance: Because co-equal ownership with practice autonomy means every partner is entitled to an opinion and none is obliged to decide. No group's revenue depends on resolving a firm-wide technology question this quarter, so it stays open, use spreads in the meantime, and each group believes its own answer is the firm's.
-
What is the difference between responsibility, accountability and authority in this lesson?
Answer guidance: Responsibility is doing the work correctly and can sit with anyone. Accountability is being answerable for the outcome by name, to someone with standing to ask, and almost always needs a licensed person. Authority is being able to say yes and no and have it stick. Governance means putting accountability and authority in the same named place.
-
What do Model Rules 5.1 and 5.3 settle, and what do they leave open?
Answer guidance: They settle that managerial and supervisory lawyers must make reasonable efforts to ensure measures are in effect giving reasonable assurance of conformity with the rules. They leave entirely open who in your firm that is. They are also model provisions; each jurisdiction adopts its own version with its own numbering, so the adopted rule is the one that binds you.
-
Does adopting NIST AI RMF 1.0 or ISO/IEC 42001 discharge a professional duty?
Answer guidance: No. Both are voluntary frameworks. They give a partnership an organising vocabulary and an agenda, and ISO/IEC 42001 can be certified by an accredited body, but adopting either neither creates nor discharges a duty under the rules of professional conduct. A firm can be fully aligned with both and still be unable to name who approved anything.
Sources and further reading
- American Bar Association — Model Rules of Professional Conduct. Accessed 2026-09-14.
- American Bar Association Standing Committee on Ethics and Professional Responsibility — Formal Opinion 512: Generative Artificial Intelligence Tools (2024-07-29). Accessed 2026-09-14.
- American Bar Association Standing Committee on Ethics and Professional Responsibility — Formal Opinion 08-451: Lawyer's Obligations When Outsourcing Legal and Nonlegal Support Services (2008-08-05). Accessed 2026-09-14.
- National Institute of Standards and Technology — Artificial Intelligence Risk Management Framework (AI RMF 1.0), NIST AI 100-1 (2023-01-26). Accessed 2026-09-14.
- International Organization for Standardization and International Electrotechnical Commission — ISO/IEC 42001:2023 Information technology — Artificial intelligence — Management system (2023-12-18). Accessed 2026-09-14.
- AI Coalition Network — Certification Educational Standard: Human Review of AI-Assisted Professional Work. Accessed 2026-09-14.