For law firms · Policy template
A written AI-use policy partners approve and associates actually read.
Copy, adapt, sign. The policy text, the Opinion 512 alignment brief, the engagement-letter consent clauses, the malpractice-carrier renewal pack, and the state-bar overlay — on this page, in editable form. No form.
Why this page exists
Two associates pasted a deposition into ChatGPT this week. The engagement letter did not mention AI.
ABA Formal Opinion 512 is a year old. The state-bar follow-ons are still landing — Florida 24-1, California Practical Guidance, DC Bar 388, NYSBA's Task Force Report, Pennsylvania 2024-200. The substance is consistent: the firm is responsible for what its associates do with AI on client confidences, the engagement letter has to satisfy a specific-consent standard, and boilerplate consent is explicitly rejected.
The firm administrator has been asking for a written policy for 18 months. The managing partner cannot draft one from scratch, map six state-bar opinions, write the engagement-letter language, and train 24 attorneys — while also running the firm.
Below is the policy. Copy it, change the bracketed values for your firm, walk it to your ethics partner. The Opinion 512 alignment brief, the engagement-letter clauses, the carrier-renewal pack, and the state-bar overlay are on this page in the same form.
Artifact 01 — The 1-page policy
Firm AI-use policy template.
Replace bracketed values with your firm's. Reads in 12 minutes. Your ethics partner reviews and approves with documented revisions. One signed copy lives in the firm's policy binder.
Policy text — copy from here
[FIRM NAME] — Generative AI Use Policy
1. Authorized tools.
The following AI tools are authorized for use on client matters: [AUTHORIZED TOOLS — e.g. Harvey, Spellbook, Casetext CoCounsel, ChatGPT Enterprise, Lexis+ AI]. No other AI tool may be used on client confidences without prior written approval from [ETHICS PARTNER]. Personal accounts on consumer AI tools (free ChatGPT, free Claude, Gemini personal) are prohibited for any client work.
2. Confidentiality (Rule 1.6).
Every authorized tool operates in a tenant-isolated environment. Inputs are not used to train the underlying model and are not accessible to other tenants of the vendor. The firm holds a written data-handling attestation from each vendor in [VENDOR FILE LOCATION]. Before any AI tool is used on a matter where client confidences are involved, the matter file must contain the specific consent clause referenced in the engagement letter (see Artifact 03 below).
3. Supervision (Rules 5.1 and 5.3).
AI is treated as nonlawyer assistance under Rule 5.3. Every AI-assisted work product is reviewed by a responsible attorney before any use, filing, or transmission. The responsible attorney is competent in the tool under Rule 1.1 (see §4 below). The supervision chain is documented in the matter file.
4. Competence (Rule 1.1, Comment 8).
Every attorney authorized to use an AI tool has completed training specific to that tool — its capabilities, its known failure modes, the verification discipline, the categories of work the tool is not used for. Training records are retained in [HR / TRAINING FILE LOCATION]. Partners who supervise AI-assisted work product are trained on the same standard as the associates.
5. Candor (Rules 3.1 and 3.3).
AI-assisted litigation work product is cite-checked against primary source before any partner signs. Standing-order compliance is verified per district before any filing. The firm maintains a current standing-order register at [REGISTER LOCATION].
6. Communication (Rule 1.4).
For every matter where AI is material to work product or fee, the matter file contains a short note from the responsible partner recording (a) that disclosure to the client was considered, (b) the decision, and (c) the basis. Where disclosure is made, the client communication is in the file.
7. Fees (Rule 1.5).
The firm does not bill clients for time the lawyer did not spend. Time spent reviewing AI-assisted work product is billable consistent with Rule 1.5. AI-assisted matters handled on alternative fee arrangements are priced consistent with the AFA terms in the engagement letter.
8. Incident response.
Any suspected confidentiality breach, hallucinated citation submitted to a tribunal, or unauthorized use of an AI tool on client confidences is reported to [ETHICS PARTNER] within [24 / 48] hours. The Ethics Partner determines whether client notification, malpractice-carrier notification, or bar reporting is required.
9. Review.
This policy is reviewed annually and after any material change to authorized tools, applicable bar opinions, or firm practice. Date of last review: [DATE]. Reviewed by: [ETHICS PARTNER NAME].
Approved by [MANAGING PARTNER NAME], Managing Partner. Reviewed by [ETHICS PARTNER NAME], Ethics Partner. [DATE].
Artifact 02 — The 1-page alignment brief
Opinion 512 → firm action, rule by rule.
The artifact the administrator forwards to the ethics partner. The artifact the ethics partner forwards to the malpractice carrier. Each row maps an operative Opinion 512 requirement to the specific section of the policy above and the operational evidence the firm produces on demand.
| Model Rule | Opinion 512 requirement | Policy § | Evidence the firm produces |
|---|---|---|---|
| 1.1 (Competence) + Comment 8 | Lawyer must understand the tool's capabilities and limitations as relevant to the matter. | §4 | Training records per attorney per tool. |
| 1.6 (Confidentiality) | Reasonable efforts to prevent unauthorized disclosure; specific consent before AI use on confidences; boilerplate rejected. | §2 | Vendor data-handling attestations; engagement-letter consent clause per matter (Artifact 03). |
| 1.4 (Communication) | Lawyer considers whether to inform the client about AI use. | §6 | Disclosure-consideration note in every AI-material matter file. |
| 3.1 / 3.3 (Candor) | Verification before any AI output reaches a tribunal. | §5 | Cite-check log per filing; standing-order register. |
| 5.1 / 5.3 (Supervision) | AI treated as nonlawyer assistance; documented supervision chain. | §3 | Responsible-attorney designation per AI-assisted work product. |
| 1.5 (Fees) | Reasonable fees; no billing for AI-saved time; review time billable. | §7 | Billing posture documented; AFA terms named in engagement letter. |
Artifact 03 — The engagement-letter clauses
Specific-consent language. Not boilerplate.
Opinion 512 explicitly rejects boilerplate. The clauses below are tool-specific and matter-type-specific. A firm with Florida exposure builds to the Florida 24-1 standard by default — informed consent before any disclosure of confidential information to a third-party AI tool. The Florida-grade clause satisfies every other current state-bar opinion at the same time.
Default clause — multi-state, Florida-grade
In connection with the Matter, we may use the following generative AI tools to assist with our representation: [AUTHORIZED TOOLS — e.g. Harvey, Spellbook, Casetext CoCounsel]. These tools may be used for [CATEGORIES OF WORK — e.g. first-draft preparation of motions and briefs, deposition-preparation summaries, contract clause comparison, legal research synthesis]. Inputs to these tools may include documents and information you provide. The tools are configured to operate in a tenant-isolated environment; inputs are not used to train the underlying AI models, are not accessible to other tenants of the vendor, and are retained only as required to support your matter. All AI-assisted work product is reviewed by a responsible attorney before any use, filing, or transmission. By signing below you specifically consent to this use. We will obtain separate written consent before using any AI tool not listed above on your matter.
Litigation matters — addendum (Rule 3.3 + standing orders)
Any AI-assisted work product filed with a tribunal is cite-checked against primary source before signature. Where the tribunal's standing order requires AI-use disclosure in filings, the firm complies with the standing order on every filing in the Matter. The firm tracks the current standing-order register for every district in which it practices.
Alternative-fee matters — addendum (Rule 1.5)
The firm does not bill for time saved by AI assistance. The fees set forth in this engagement letter reflect the value of the work product delivered, not time spent. Where AI is material to the work product, the firm retains internal records of AI-assisted activity sufficient to satisfy Rule 1.5's reasonableness standard on review.
A starting-point sample. Not legal advice. Your ethics partner reviews and approves before deployment. The clauses are written to the highest current state-bar standard (Florida 24-1) so a single clause covers a multi-state footprint.
Artifact 04 — The carrier-renewal pack
The single document the broker reads instead of asking the governance question.
At least one major program now excludes claims attributable to generative AI use. The carrier renewal questionnaire this cycle asks specifically what governance the firm has in place. The pack below is the document you hand the broker before the call. Section structure follows the typical AI-governance section of a renewal questionnaire.
Carrier-renewal pack — section outline
A. Written AI-use policy.
Attached. Approved by Managing Partner and Ethics Partner. Date of last review: [DATE].
B. Authorized tools and vendor data-handling.
The firm authorizes [N] tools, listed in §1 of the policy. For each tool, the firm holds a written vendor attestation covering tenant isolation, no training on inputs, retention policy, and incident-response posture. Available on request.
C. Engagement-letter language.
Specific-consent language is included in every engagement letter for matters where AI is material. Built to Florida Bar 24-1 standard. Sample clause attached (Artifact 03).
D. Supervision protocol.
§3 of the policy. Responsible-attorney designation per AI-assisted work product. Documented in matter file.
E. Training program.
§4 of the policy. Training is tool-specific and applies to partners and associates equally. Records retained in [LOCATION].
F. Litigation verification.
§5 of the policy. Every AI-assisted filing cite-checked against primary source. Standing-order register current as of [DATE].
G. Incident response.
§8 of the policy. Reporting timeline: [24 / 48] hours. No reportable incidents in the current policy period. [OR — Incidents reported and addressed: [SUMMARY], with remediation.]
Artifact 05 — The state-bar overlay
Where each jurisdiction is stricter than Opinion 512.
A multi-state firm builds to the highest standard once and satisfies every jurisdiction in its footprint. The Florida-grade engagement-letter clause satisfies every other state currently issuing AI guidance. The disclosure default in NYSBA is the higher bar a firm with New York exposure should adopt.
| Jurisdiction | Document | Where it goes beyond Opinion 512 | Operational impact |
|---|---|---|---|
| Florida | Florida Bar Ethics Opinion 24-1, January 2024 | Informed consent before any disclosure of confidential information to a third-party generative AI program. | Engagement-letter clause obtains specific consent before any AI use on confidences. The default for any multi-state firm. |
| California | California State Bar Practical Guidance, November 2023 | Explicit billing posture: cannot bill time saved by AI; can bill for review time. | Policy §7 names the billing posture; engagement letters reflect. |
| District of Columbia | DC Bar Ethics Opinion 388, April 2024 | Consolidates Opinion 512 framework; emphasizes supervision documentation. | Policy §3 supervision chain is the operational answer. |
| New York | NYSBA Task Force on AI Report, April 2024 | Most disclosure-forward state guidance; treats client disclosure of AI use as the default. | Policy §6 treats disclosure as default in AI-material matters; matter file documents. |
| Pennsylvania | Pennsylvania / Philadelphia Bar Formal Opinion 2024-200 | Mirrors Opinion 512 with practitioner examples; emphasizes Rule 1.1 training. | Policy §4 training records satisfy. |
| New Jersey | NJ Supreme Court Notice on AI, January 2024 | Practitioner guidance — verification of citations a Rule 3.3 obligation. | Policy §5 cite-check protocol satisfies. |
Read or forward
Five formats. Same five artifacts.
Word template (.docx)
The editable policy
The policy, the engagement-letter clauses, and the carrier-renewal pack — in one editable Word document. Placeholders highlighted in yellow. Open in Word, replace, save under your firm's filename.
Download .docx →Article
Read on this page
The policy, the alignment brief, the consent clauses, the carrier pack, and the state-bar overlay — all five artifacts, top to bottom.
Scroll up →Full PDF
The 35-page reading version
Footnoted, source-linked, print-formatted. The version the ethics partner reviews. All five artifacts, primary-source citations, operative-language excerpts, and the adoption checklist.
Download →Summary PDF
The 2-page forwarder
Single sheet, both sides. The 1-page policy on the front, the 1-page alignment brief on the back. The version the administrator forwards to the managing partner.
Download →Slideshow PDF
The 14-slide partner-meeting deck
Designed for a 12-minute walkthrough at the next partners meeting. Slide-per-rule structure with the carrier pack at the close.
Download →No form, no email gate. PDFs forward freely.
The math the administrator brings to the partners meeting
Insurance math, not productivity math.
The cost of one avoided malpractice claim — typically $500K-$5M in settlement and premium impact, per Lawyers Mutual and Aderant research — exceeds the engagement fee for the full Ethics-Alignment Diagnostic by an order of magnitude. The firm does not need to avoid every claim. It needs to avoid one.
The policy on this page is a starting point. The Ethics-Alignment Diagnostic is the engagement that audits the firm's actual AI use, calibrates the policy to your tools and practice areas, drafts the engagement-letter language for your client mix, and trains your attorneys.
Cost of one avoided malpractice claim
$500K–$5M
Lawyers Mutual / Aderant research, 2024.
Ethics-Alignment Diagnostic, fixed fee, 4 weeks
$15K
One avoided claim funds the engagement 33-333×.
Operative Model Rules in Opinion 512
6
Rules 1.1, 1.4, 1.5, 1.6, 3.3, 5.1/5.3 — every rule the policy maps to.
Three quotes from the regulators
Pulled directly. Attributed precisely.
Boilerplate consent included in engagement letters will not be adequate when AI tools involve disclosure of client confidences.
— ABA Formal Opinion 512, July 29, 2024
A lawyer must obtain the affected client's informed consent prior to utilizing a third-party generative AI program if the utilization would involve the disclosure of any confidential information.
— Florida Bar Ethics Opinion 24-1, January 2024
Lawyers using GAI tools must maintain technological competence, protect confidentiality, communicate with clients about the use of GAI, supervise GAI as a nonlawyer assistant under Rule 5.3, ensure candor toward the tribunal, and bill in a manner consistent with Rule 1.5.
— ABA Formal Opinion 512, summary obligations
After the policy is signed
The defendable line, in the managing partner's vocabulary.
The ethics partner has reviewed and approved the policy. The engagement letters satisfy Opinion 512's specific-consent standard. The malpractice carrier has the governance summary on file. The associates are trained on the tools and the ethical guardrails. The firm administrator owns the policy and the rollout after the diagnostic concludes.
When the template is not enough on its own
Have us run the diagnostic for your firm.
You describe the situation — the practice areas, the tools already in use, the malpractice carrier's questions at the last renewal. We audit your actual AI use, calibrate the policy to your tools and engagement types, draft the engagement-letter language for your client mix, and train your attorneys. Fixed fee. Four weeks at most firms. The diagnostic stands on its own if you stop after it.
Direct: todd@zusmanpartners.com.