# Case: The Firm That Added an AI Clause to Every New Contract **By Dan Cumberland** · Published May 7, 2026 · Categories: AI Strategy > AI contract drafting crossed into the mainstream in 2025. According to the Thomson Reuters Institute[^1], 26% of legal organizations are now actively using... ## Why AI Contract Drafting Crossed the Line This Year AI contract drafting crossed into the mainstream in 2025\. According to the Thomson Reuters Institute[1](/blog/blog-ai-contract-drafting#ref-1), 26% of legal organizations are now actively using generative AI— up from 14% the year prior— and 76% of in\-house legal professionals identify contract drafting and review as the top use case\. The same report found 78% of law firm respondents expect generative AI to become central to their workflow within five years\. That is the demand side\. The supply side caught up too\. Goldman Sachs[2](/blog/blog-ai-contract-drafting#ref-2) estimated that 44% of legal tasks could be automated by AI— a number that gets quoted often, and for good reason\. The work is automatable\. The duty is not\. ```html-table
What Changed20242025
Active gen AI use in legal orgs14%26%
Top in-house use case(varied)Contract drafting (76%)
Authoritative ethics guidancePatchworkABA Formal Op. 512 (July 2024)
State-level transparency lawNone bindingCA SB 942 effective Jan 1, 2026
``` ABA Formal Opinion 512[3](/blog/blog-ai-contract-drafting#ref-3) \(issued July 29, 2024\) made the duty floor explicit: lawyers using generative AI must maintain competence \(Rule 1\.1\), protect confidential client data \(Rule 1\.6\), and verify AI outputs before they reach a client\. And the regulatory tailwind is real\. The California AI Transparency Act[4](/blog/blog-ai-contract-drafting#ref-4) \(SB 942\), effective January 1, 2026, requires covered providers with over one million monthly users to offer detection tools, watermark generated content, and publish training\-data summaries\. That pushes disclosure norms downstream into vendor and client contracts whether firms want it or not\. With the duty floor set, the question is what to actually put in the clause\. ## What the Clause Actually Said \(Annotated\) An AI clause should specify four things: scope of AI use, data restrictions, principal review, and billing treatment\. In that order\. Below is the paragraph the firm adopted, adapted from a practitioner template\. > *Illustrative only; consult counsel before adopting\.* "The Firm may use artificial intelligence\-assisted technology as a supplementary tool for legal research, drafting, proofreading, and administrative support[5](/blog/blog-ai-contract-drafting#ref-5)\. No confidential or client\-identifying information will be entered into any general\-purpose, public AI tool; only enterprise tools with appropriate data\-handling protections will be used for client matters\. All AI\-assisted output is reviewed and approved by an attorney before delivery to the Client; the Firm— not the tool— is responsible for the work\. The Firm bills for attorney time on the matter and does not bill the cost of the AI tool itself unless agreed in writing in advance\." Each sentence does specific work\. And each one ties back to a duty\. ```html-table
SentenceWhat It DoesABA Duty Implicated
Scope sentenceNames what AI is used for— research, drafting, proofreadingCompetence (Rule 1.1)
Data sentenceFences confidential information out of public LLMsConfidentiality (Rule 1.6)
Review sentenceAssigns the human reviewer and locates responsibilityVerification (Op. 512)
Billing sentenceRemoves ambiguity on who pays for the toolReasonable fees (Rule 1.5)
``` A workable AI clause names the scope, fences the data, assigns the reviewer, and clarifies billing\. Practitioner templates from Taft[6](/blog/blog-ai-contract-drafting#ref-6) add IP, security, and termination language for higher\-stakes matters\. Open standards from Bonterms[7](/blog/blog-ai-contract-drafting#ref-7) cover the cross\-industry case\. One more distinction worth making\. Disclosure tells the client AI is being used\. Consent asks the client to authorize it\. Many matters can run on disclosure alone\. Some— particularly anything touching regulated data, M&A, or litigation work product— move from notice to affirmative consent\. The clause should be clear which one it is doing\. A clause is a promise\. The harder part is the operational discipline that makes the promise true\. ## What the Clause Forced Inside the Firm The clause is the easy part\. The hard part is the four operational rules the firm now has to keep: a confidential\-data fence, a principal verification step, a tool\-approval list, and a no\-AI\-surcharge billing rule\. A clause without an operational discipline behind it is a liability waiting to happen\. Here is what the firm actually had to build, and where the [AI strategy roadmap](/services/ai-strategy/) work shows up: 1. **The data fence\.** Decide which inputs never go into a public LLM\. Client identities, matter facts, draft work product, and anything covered by a confidentiality obligation belong inside enterprise tools with contractual data\-handling protections— not free\-tier consumer chatbots\. 2. **The verification step\.** Every AI\-assisted output gets reviewed by an attorney \(or, for non\-law firms, the responsible principal\) before it leaves the building\. This is the step ABA Op\. 512 effectively requires, and it is the one that erases AI's time savings if the workflow is sloppy\. 3. **The tool\-approval list\.** The firm picks which tools are in bounds— illustratively, things like Spellbook, Harvey, Lexis\+ with Protégé, or CoCounsel— and which are not\. Free\-tier ChatGPT for client work is not on the list\. Period\. 4. **The billing rule\.** Bill for the human time on the matter\. Do not pad bills for AI\. If a matter genuinely required a paid tool the firm is passing through, say so in writing in advance\. The verification step is where the math gets honest\. A 2018 LawGeex study[8](/blog/blog-ai-contract-drafting#ref-8), vendor\-funded but illustrative, found 20 experienced lawyers averaged 92 minutes to review five NDAs while AI completed the same task in 26 seconds\. That is the upside\. Verification overhead is what keeps it safe\. Just because it is easy does not mean it is good— the discipline is what turns easy into good\. All of this is internal\. The interesting part is what happened outside\. ## The Trust Outcome \(and Why Clients Did Not Freak Out\) Disclosed AI use, when the clause is written well, functions as a competence signal— not a liability\. Clients respond to clarity, not vagueness\. > Clients are not afraid of AI\. They are afraid of being surprised by it\. The Reddit fear— "will clients freak out?"— assumes silence is safer than disclosure\. It is the opposite\. The Thomson Reuters 2025 data[1](/blog/blog-ai-contract-drafting#ref-1) showed legal\-professional hesitancy on generative AI dropped to 24% from 35% the year prior, which is at least a directional proxy: the people closest to the work got more comfortable with disclosure, not less\. The firms that operationalized AI early are using their clauses in pitches and renewal conversations\. The clause moves the conversation from *are you using AI?* to *how are you using AI to deliver more for me?* Honest caveat: no primary client\-side survey data exists yet on how clients react to receiving an AI clause— the trust argument here is positioning, not measurement\. But silence while using AI anyway is the move that actually carries risk\. The clause is not what creates the exposure; the silence is\. This is not a law\-firm\-only move\. ## Where Founder\-Led Firms Outside Law Should Copy This AI contract drafting is not a law\-firm\-only practice\. Any founder\-led firm using AI in client\-facing work should consider an AI clause in their service agreements\. If running an agency, consultancy, or accounting practice, the same four ingredients apply— translated into the language of the work\. - **Agencies** \(AI in copy, design, research\) → disclosure clause \+ named human reviewer\. Clients are paying for judgment, not AI output\. - **Consultancies** \(AI in analysis, deck generation, synthesis\) → reviewer assignment \+ data\-handling fence for client information\. - **Accounting and bookkeeping firms** \(AI in tax research, classification, reconciliation\) → confidentiality fence is the non\-negotiable; nothing client\-identifying goes into a public model\. Open standards like Bonterms[7](/blog/blog-ai-contract-drafting#ref-7) make this portable\. And the Baker Donelson 2026 forecast[4](/blog/blog-ai-contract-drafting#ref-4) flagged that compliance norms are pushing AI disclosure language downstream into ordinary commercial contracts, not just professional\-services ones\. The framing that holds this together is one Dan Cumberland Labs uses with [founder\-led firms](/for-founders/) building their AI operating posture: AI is intellectual augmentation, not artificial intelligence\. The firm's judgment is still the product\. The clause exists because that is still true\. If using AI in client work and the contract is silent, the firm is gambling that no client will ever ask\. ## FAQ ### Do lawyers have to tell clients they used AI? Not always required by rule, but ABA Formal Opinion 512[3](/blog/blog-ai-contract-drafting#ref-3) \(July 2024\) requires lawyers to maintain competence, protect confidential client data, and verify AI outputs\. Many firms now disclose proactively to align with these duties and to reduce client surprise\. Disclosure is the safer default once AI is part of the workflow\. ### What should an AI clause include? At minimum: scope of AI use, data\-handling restrictions, attorney \(or principal\) verification, and billing treatment\. Practitioner templates from Ellis[5](/blog/blog-ai-contract-drafting#ref-5) and Taft[6](/blog/blog-ai-contract-drafting#ref-6) add IP, security, and termination language\. The Bonterms open standard[7](/blog/blog-ai-contract-drafting#ref-7) covers cross\-industry adaptation\. ### Can I bill clients for AI\-assisted work? Yes— for the attorney's or professional's time on the matter[5](/blog/blog-ai-contract-drafting#ref-5)\. The cost of the AI tool itself is generally not billed separately unless agreed in writing in advance\. Padding bills for AI use that displaced human hours is the practice that gets firms in trouble\. ### What is the difference between AI disclosure and AI consent? Disclosure tells the client AI is being used\. Consent asks the client to authorize it\. Some matters and jurisdictions move from notice toward affirmative consent[3](/blog/blog-ai-contract-drafting#ref-3)— the clause should be clear which one it is doing for any given engagement\. ### Does an AI clause protect the firm legally? A clause documents the firm's standard of care and the client's awareness of AI use, but it does not waive professional\-responsibility duties[3](/blog/blog-ai-contract-drafting#ref-3)\. Operational discipline— data fences, verification, approved tools— is what actually reduces risk\. The paragraph is the promise; the workflow is the protection\. ## Conclusion An AI clause is the smallest contract change with the largest signaling effect\. Done well, it tells clients the firm is competent, transparent, and modern— and it forces the firm to be all three\. The clause is one paragraph\. The discipline behind it is the firm's whole AI operating posture\. If mapping AI policy across client work, tools, and contracts feels like a full\-time job on its own, that is exactly the kind of problem an [AI implementation services](/services/ai-implementation/) partner can compress\. [Dan Cumberland Labs](https://dancumberlandlabs.com) helps founder\-led firms build the operational discipline behind the paragraph— so the clause becomes a true statement, not a wish\. ## References 1. Thomson Reuters Institute, "2025 Generative AI in Professional Services Report" \(2025\) — [https://www\.lawnext\.com/2025/04/thomson\-reuters\-survey\-over\-95\-of\-legal\-professionals\-expect\-gen\-ai\-to\-become\-central\-to\-workflow\-within\-five\-years\.html](https://www.lawnext.com/2025/04/thomson-reuters-survey-over-95-of-legal-professionals-expect-gen-ai-to-become-central-to-workflow-within-five-years.html) 2. Goldman Sachs \(via Law\.com\), "Generative AI Could Automate Almost Half of All Legal Tasks, Goldman Sachs Estimates" \(2023\) — [https://www\.law\.com/legaltechnews/2023/03/29/generative\-ai\-could\-automate\-almost\-half\-of\-all\-legal\-tasks\-goldman\-sachs\-estimates/](https://www.law.com/legaltechnews/2023/03/29/generative-ai-could-automate-almost-half-of-all-legal-tasks-goldman-sachs-estimates/) 3. American Bar Association, "ABA Formal Opinion 512: Generative Artificial Intelligence Tools" \(2024\) — [https://www\.americanbar\.org/groups/professional\_responsibility/publications/formal\_opinions/](https://www.americanbar.org/groups/professional_responsibility/publications/formal_opinions/) 4. State of California \(via CPO Magazine\), "2026 AI Legal Forecast: From Innovation to Compliance" \(2026\) — [https://www\.cpomagazine\.com/data\-protection/2026\-ai\-legal\-forecast\-from\-innovation\-to\-compliance/](https://www.cpomagazine.com/data-protection/2026-ai-legal-forecast-from-innovation-to-compliance/) 5. J\. Lee Ellis, "Sample Artificial Intelligence Clauses for Attorney Engagement Agreements" \(2025\) — [https://jlellis\.net/blog/sample\-artificial\-intelligence\-clauses\-for\-attorney\-engagement\-agreements/](https://jlellis.net/blog/sample-artificial-intelligence-clauses-for-attorney-engagement-agreements/) 6. Taft Stettinius & Hollister LLP, "The Expanding Prevalence of AI Clauses in Contracts" \(2025\) — [https://www\.taftlaw\.com/news\-events/law\-bulletins/the\-expanding\-prevalence\-of\-ai\-clauses\-in\-contracts/](https://www.taftlaw.com/news-events/law-bulletins/the-expanding-prevalence-of-ai-clauses-in-contracts/) 7. Bonterms, "AI Standard Clauses Version 1\.0" \(2025\) — [https://bonterms\.com/forms/ai\-standard\-clauses\-version\-1\-0](https://bonterms.com/forms/ai-standard-clauses-version-1-0) 8. LawGeex \(via Virtasant\), "AI Contract Management: 80% Time Savings in Legal Work" \(2018\) — [https://www\.virtasant\.com/ai\-today/ai\-contract\-mangement\-legal](https://www.virtasant.com/ai-today/ai-contract-mangement-legal) --- Source: https://dancumberlandlabs.com/blog/ai-contract-drafting/