Billing for AI-Assisted Drafting: How Outside-Counsel Guidelines Are Catching Up to Rule 1.5

The American Bar Association's Formal Opinion 512, issued July 29, 2024, told the profession what it half-knew: generative drafting tools engage the duties of competence, confidentiality, supervision, and — pointedly for the bills — fee reasonableness under Model Rule 1.5, so a lawyer who produces a brief in two hours with machine assistance cannot bill ten merely because ten hours used to be the price. Since then, the operational question has migrated from the ethics opinion into outside-counsel guidelines, where corporate legal departments are writing the same principle into engagement letters as deliverables and pricing terms.

3G Times publishes information, not legal advice; professional-responsibility questions turn on jurisdiction and engagement facts, and state analogues of the Model Rules vary.

What Opinion 512 actually settled

The opinion runs the generative-AI toolset across the familiar rule set. Competence requires understanding the tools well enough to use them responsibly — including their tendency to fabricate citations, which ties back to candor toward tribunals. Confidentiality requires informed consent before client information enters self-learning tools that retain inputs; supervision extends Rule 5.3's nonlawyer-assistance logic to vendors and machine outputs alike. On fees, the opinion is brief but categorical: billing must reflect the work actually done, and reclaimed hours are not a billing opportunity. The opinion did not prescribe disclosure to clients in every case — it left disclosure to be judged against the client's reasonable expectations, which is precisely the gap engagement terms now fill.

Why the billing question moved to the engagement letter

Rule 1.5's reasonableness standard is a discipline enforced after the fact, by bars and fee disputes. Buyers of legal services prefer enforcement before the fact: if the department's metric is cost per matter, an associate's AI-accelerated first draft billed at legacy hours is a margin transfer from client to firm, not a fee dispute in waiting. Hence the direction of travel in OCG revisions: disclosure of generative-AI use in the matter, no billing of machine-generated drafting absent agreed terms, pass-through charges for AI tools flagged as disbursements, and — increasingly — task codes that distinguish AI-assisted work so the blended rate can be argued honestly in both directions. None of this contradicts the opinion; it prices it.

What outside-counsel guidelines now ask for

The table's quiet theme is that every clause is an ethics duty rendered auditable. A department that receives tool names and task-code splits can reconcile invoices against its own understanding of the work; one that receives only "AI may be used" has bought the disclosure without the accounting.

Audit mechanics complete the picture. An OCG clause without an invoice-code reconciliation is a statement of values, not a control; the audit provision should give the department the right to map AI-flagged task codes to billed hours in the annual review, and the firm a clean path to explain anomalies before they become disputes. Departments that schedule the reconciliation routinely report that the disclosure question moves from suspicion to bookkeeping.

How firms are pricing the same reality

Seller-side responses have split along predictable lines. Some firms hold blended rates and absorb the technology cost, treating reclaimed hours as the subsidy that pays for the platform — a position easy to state and hard to audit. Others move affected tasks to fixed or subscription fees, which converts the efficiency into a shared, visible number. The minority position — bill as before, disclose nothing until asked — is the one the ethics opinion forecloses, and the one buyers now write clauses against. For fintech legal teams that live in regulatory-response work, the fixed-fee drift matters most: drafting cycles that were billed as iterations compress into a single reviewed product, and the fee should name that compression rather than inherit the old cadence.

What have the state bars added since?

The ABA's opinion set the frame; the states filled the corners. Florida's Professional Ethics Opinion 24-1, adopted in June 2024, ran ahead on specifics — cautioning that client data must not be disclosed to generative tools without informed consent, that lawyers may not charge for time saved by the technology, and that a chatbot's output is the filing lawyer's work product in every sense that matters. Through 2024 and 2025, bar after bar issued similar guidance in the same shape: confidentiality gating, supervision duties, citation verification, and billing reasonableness, each mapped to the jurisdiction's own rules. Two consequences follow for engagement-letter drafting. First, multi-jurisdictional matters cannot assume one ethics regime — the OCG's AI provisions should be drafted to the strictest applicable standard rather than the forum's, which is how sophisticated departments already handle privilege conflicts. Second, state guidance tends to move faster than the ABA on specific tools and practices, so the OCG's change-notice duty should extend to the ethics landscape: a clause requiring the firm to flag new bar guidance affecting its AI workflow keeps the contract current without renegotiation.

What does this mean in practice?

• Put the tool list in the OCG with a change-notice duty. Disclosure obligations that name products survive vendor switching; generic acknowledgments do not.
• Ask for task-code granularity before rate concessions. The honest argument about AI-assisted bills needs the split between machine draft and lawyer hours; negotiate the codes first, the percentages second.
• Mirror the confidentiality clause to the data. Regulatory-response matters carry filing-timing sensitivity; no-training and jurisdiction-routing terms should track that exposure, not the firm's default.
• Keep the verification warranty mutual. The client-side benefit of speed dies if every draft returns for full human re-verification; the clause should allocate that labor deliberately.

Opinion 512 will not be the last word — state bars continue to issue their own guidance, and court rules on AI disclosures will sharpen it. But the economic settlement is already visible: the hours a machine saves have become a term of trade, and the parties writing it down first are the ones who keep the margin they intended.

Frequently asked questions

Can a firm bill AI-tool subscriptions to the client?

As a disbursement, yes, where the engagement says so and the cost basis is shown. Surprise line items fail Rule 1.5 communication duties in spirit and most OCG audit provisions in letter; the clean pattern is a stated technology charge agreed at engagement.

Does Opinion 512 require telling the client whenever AI is used?

Not categorically. It requires evaluating client expectations and obtaining informed consent where confidentiality is implicated — self-learning tools fed client data. Departments that want matter-level disclosure regardless get it by contract, which is the OCG's job.

What is the citation-failure exposure?

Fabricated citations remain the duty with teeth: competence and candor make human verification non-delegable, and courts have sanctioned filers regardless of the tool involved. The OCG warranty allocates that verification cost; it cannot waive the duty.