The AI Learning Hub Journal

Billing, Consent, and Malpractice

The commercial and liability edges — questions, not answerseach is a decision the firm makes in policy; none of it is advice on your obligationsBILLINGWhen the work takes far lesstime than it used toHours billed fall; the value delivereddoes not. Fees have to stayreasonable, and reasonableness isjudged on more than a clock.QUESTIONS THE POLICY MUST ANSWERDo you bill time, or value?Who benefits from the time saved?Are tool costs recovered, and how?Is any of this visible to the client?CONSENT AND TRANSPARENCYWhat the client is told,and when they are told itEngagement terms are where tool useis either disclosed or left implied.Some clients will have their ownrules, and those rules win.QUESTIONS THE POLICY MUST ANSWERDoes the engagement letter say?Is consent asked for, or assumed?Which matters need tighter limits?Who checks client-side restrictions?MALPRACTICE EXPOSUREWhere liability sits whenunverified output goes outThe tool has no duty to the clientand cannot be sued for the advice.Exposure stays with the people andthe firm that put a name on it.QUESTIONS THE POLICY MUST ANSWERWho signs off before it leaves?Is the review step recorded?Does the insurer know the practice?What happens when it goes wrong?THESE ARE POLICY QUESTIONS, NOT SETTLED ANSWERSThere is no single correct answer to any of them — but a firm that has not answered has answered by defaultHow the underlying duties apply is jurisdiction-specific; treat this as a checklist for your own policy workThe efficiency arrives before the policy does — which is exactly why the questions need answering first
The efficiency arrives before the policy does — these are questions a firm has to answer for itself, not answers this course can give

Billing When AI Compresses the Hours

A common principle across jurisdictions is that fees must be reasonable and that a client should not be charged for time not spent. AI creates an obvious tension in hourly billing: if a memo that took six hours now takes two, billing six is difficult to defend on any reading of that principle. Billing two is straightforward but transfers the entire efficiency gain to the client while the firm carries the tool cost, the training cost, and the verification burden. There is no settled answer and practice is genuinely varied — some firms bill actual time, some shift affected work to fixed fees, some price by outcome. What is not defensible is billing time that was not worked, and that much is clear regardless of the model chosen.

  • Reasonableness of fees and not charging for unspent time are widely shared principles
  • Billing pre-AI hours for AI-compressed work is difficult to defend under either principle
  • Practice varies genuinely — actual time, fixed fees, and outcome pricing are all in use
  • The clear line is that time not worked cannot be billed as time worked

Charging for the Tools

A separate question is whether AI costs can be passed to clients as a disbursement. General principles about disbursements — that they should reflect actual cost, be disclosed, and not be a hidden profit centre — apply here as they do elsewhere, though the specifics differ by jurisdiction. Practical complications: subscription tools do not have a natural per-matter cost, allocating a firm-wide licence across matters is arbitrary, and clients are increasingly asking directly about AI use and pricing in engagement negotiations. The workable approach is to decide the position deliberately, state it in the engagement letter, and be able to explain the basis. Improvised line items discovered by a client during a fee review tend not to end well.

  • Disbursement principles apply: actual cost, disclosed, not a concealed margin
  • Subscription tools resist per-matter allocation, which makes any split arbitrary and worth explaining
  • Clients are asking about AI use and pricing directly in engagement negotiations
  • Decide the position, put it in the engagement letter, and be able to justify the basis

Client Consent and Engagement Language

Whether clients must be told about AI use is unsettled and jurisdiction-dependent. Arguments for disclosure include that clients may care, that confidentiality of their material is affected, and that trust is cheaper to maintain than to rebuild. Arguments against include that firms do not itemise every tool and that the duty is to competent work rather than to a particular method. A reasonable default is proportionality: routine internal use of approved tools sits alongside other work methods, while use that sends client material to a third party, or that materially shapes advice, warrants explicit engagement-letter language. Note also that clients increasingly impose their own AI restrictions in outside counsel guidelines, which must be tracked per client.

  • Disclosure obligations are unsettled and vary — do not assume either answer is universal
  • Proportionality is a defensible default: disclose where material leaves the firm or advice is shaped
  • Engagement-letter language beats a conversation nobody recorded
  • Client-imposed AI restrictions in outside counsel guidelines must be tracked per client

Malpractice Exposure

The malpractice picture has two sides and firms usually consider only one. Using AI badly creates exposure: a fabricated citation, a missed clause, a breach of confidentiality, an error the practitioner could not explain because they did not understand the tool. But not using AI may eventually create exposure of its own, if a standard of care develops in which failing to run an available check on a large document set becomes unreasonable. That standard is not settled and should not be overstated. Two practical steps in the meantime: confirm with your insurer how AI-assisted work is treated under your policy, because exclusions and notification conditions are appearing, and keep records that show the method you actually used.

  • Exposure from bad use is immediate and documented; exposure from non-use is speculative and slower
  • A standard of care may develop around available checks, but it is not settled — do not overstate it
  • Confirm with your insurer how AI-assisted work is treated; exclusions and conditions are appearing
  • Keep records evidencing the method used — reconstruction after a claim is much harder

Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.