Short answer: ABA Formal Opinion 512 explains how lawyers using generative AI must consider competence, confidentiality, communication, meritorious claims, candor, supervision and fees. Before use, assess the tool and confidentiality risks; during the work, apply appropriate verification and supervision; before filing or billing, review court-bound outputs and check the applicable court and fee rules.
This checklist is for practicing US lawyers. In the UK, Australia and India, local prices, taxes and professional rules differ and must be checked locally. For broader tool-selection guidance, see AI for lawyers.
What does ABA Formal Opinion 512 require?
It applies existing professional duties to generative AI use. The ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512 on July 29, 2024.
Under Model Rule 1.1, lawyers need a “reasonable understanding of the capabilities and limitations” of the tool, not expertise. The level of independent verification depends on the tool and the task.
Confidentiality requires a separate assessment. Before entering information relating to a representation, evaluate whether others inside or outside the firm could access it. Self-learning tools raise the risk of one client’s information surfacing in another matter; informed client consent is required before entering representation information into such a tool.
Idea generation without information relating to the representation does not require client informed consent. That exception does not remove the need to understand the tool or assess its output.
What should I check before, during and after AI use?
Work through the duties below at the relevant stage of the task. The actions translate Opinion 512 into a workflow; suggested notes and records are documentation aids, not proof that a duty has been satisfied. California-specific items come from its 2026 Practical Guidance.
Competence: before choosing or using a tool
-
Read the tool’s Terms of Use, privacy policy and related contractual terms, or consult someone who has. Opinion 512 says all lawyers should do this (Rules 1.1 and 1.6).
-
Assess the tool’s capabilities and limitations for the proposed task. Consider recording that assessment, but do not treat a written note as sufficient by itself to establish reasonable understanding (Rule 1.1).
-
Set an independent-verification approach appropriate to the tool and task. Do this before relying on the output; court submissions require review of outputs, including citations (Rule 1.1; Rules 3.1 and 3.3).
-
Retain control of strategic decisions. California’s guidance requires lawyers to review, verify and correct AI outputs while keeping control of strategy (California, Rules 1.1 and 1.3).
Confidentiality: before entering representation information
-
Evaluate the risk that information could be disclosed to, or accessed by, others inside or outside the firm before entering it into a GAI tool (Rule 1.6).
-
Assess whether the tool is self-learning and how it handles inputs. Review the relevant terms rather than assuming data handling from a product label. California’s AI resources warn that products may use inputs to train or refine models (Rule 1.6; California, Bus. & Prof. Code §6068(e)).
-
Obtain informed client consent before entering representation information into a self-learning tool. Opinion 512 says “merely adding general, boiler-plate provisions to engagement letters” is insufficient (Rule 1.6).
-
Keep consent-free idea generation free of information relating to the representation. The opinion’s exception depends on that condition (Rule 1.6).
-
Apply the confidentiality assessment to packaged workflows too. Before using a workflow such as our NDA review skill for lawyers, evaluate the underlying tool’s terms and disclosure risks (Rule 1.6).
Communication: when planning the work
-
Decide whether the client needs to be told about GAI use. Rule 1.4 may require communication even where Rule 1.6 does not require informed consent; consider recording your reasoning (Rule 1.4).
-
Consider disclosure of intended AI use, how it will be used, and its benefits and risks if practicing in California (California, Rules 1.2 and 1.4).
Meritorious claims and candor: before filing
-
Review outputs, including citations, before submitting them to a court. A generated citation belongs in the review queue, not automatically in the final filing (Rules 3.1 and 3.3).
-
Consult the court’s local rules on AI use. Opinion 512’s footnote 47 directs lawyers to those rules (Rule 3.3).
-
Retain responsibility for candor. California’s guidance says the duty cannot be delegated to AI and requires review of all AI outputs for accuracy before submission (California, Rules 3.1 and 3.3).
Supervision: throughout the workflow
-
Set firm policies and supervise lawyers and nonlawyers if you have managerial responsibilities. A written policy is a practical way to communicate expectations (Rules 5.1 and 5.3).
-
Train the people using GAI. Opinion 512 includes training within supervision; California specifies periodic AI training (Rules 5.1 and 5.3; California, Rules 5.1–5.3).
-
Require meaningful lawyer supervision of agentic systems making substantive legal determinations or filing pleadings. California says lawyers must not deploy those systems for these tasks without it (California, Rules 5.1–5.3).
Fees: before invoicing
-
Reassess a flat fee when GAI makes the task much faster. Opinion 512 says charging the same flat fee in that situation may be unreasonable (Rule 1.5).
-
Exclude general office overhead from client charges. Opinion 512 says it may not be billed (Rule 1.5).
-
Check the conditions for any AI-cost surcharge. Opinion 512 prohibits a surcharge on actual GAI expense absent disclosure; California requires informed written consent for a markup (Rule 1.5; California, Bus. & Prof. Code §§6147–6148).
-
Record actual time for hourly billing. California permits charges for time actually spent crafting prompts and reviewing outputs, and hourly entries must reflect time actually spent (California, Rule 1.5).
The table summarizes the duties and suggests evidence to retain. These records do not replace the underlying assessment or any applicable consent requirements.
Updated 2026-10-20. Based on ABA Formal Opinion 512 and California’s 2026 Practical Guidance.
| Duty | Rule | What to do | Evidence to keep |
|---|---|---|---|
| Competence | 1.1 | Understand capabilities and limits; choose appropriate verification | Terms reviewed; task assessment; verification plan |
| Confidentiality | 1.6 | Assess disclosure risk; obtain informed consent for representation information in self-learning tools | Risk assessment; record of specific informed consent |
| Communication | 1.4 | Assess whether the client must be told | Decision note; client communication |
| Meritorious claims and candor | 3.1, 3.3 | Review outputs and citations; consult local AI rules | Completed review checklist; local-rule check |
| Supervision | 5.1, 5.3 | Set policies, supervise and train | Policies; training records; review assignments |
| Fees | 1.5 | Review flat fees, overhead and surcharges; record actual hourly work under California guidance | Time records; cost disclosures; fee terms |
How should I verify citations before filing?
Review the actual authorities and whether they support the propositions in your filing; use AI only to organize the checklist below. Opinion 512 requires review of outputs, including citations, before court submission, and California says candor cannot be delegated to AI.
As a practical workflow, check the source text, citation details, relevant passage and subsequent treatment yourself. A model’s assertion that an authority exists does not complete those checks.
Apply the confidentiality checklist before using this prompt. Citations and supporting propositions can themselves relate to a representation; removing names alone does not resolve that issue. Use fictional examples unless the required risk assessment and any necessary consent are in place. Adapt other workflows from our prompt library for lawyers with the same caution.
You are helping a licensed attorney organize a pre-filing citation-verification checklist. Do not invent, guess, reconstruct or substitute legal authorities.
Use fictional material unless the attorney has completed the necessary confidentiality assessment and obtained any required consent. Do not request client names, confidential facts or privileged analysis.
Document type: [DOCUMENT TYPE]
Court or tribunal: [COURT OR TRIBUNAL]
Jurisdiction: [JURISDICTION]
Material approved for this tool:
[CITATION] supports [PROPOSITION]
[ADDITIONAL CITATIONS AND PROPOSITIONS]
For each citation, create a row with:
- Citation exactly as supplied
- Proposition exactly as supplied
- Checks for the attorney: existence; caption, court and year; reporter and pinpoint reference; support for the proposition in the source text; subsequent history and treatment; current validity
- Status: UNVERIFIED
Instructions:
- Mark every citation UNVERIFIED.
- Never claim that an authority exists, is accurate or remains good law.
- Do not add, suggest or substitute authorities.
- Flag malformed citations as "format check needed"; do not repair them.
- If asked to supply an authority, decline and direct the attorney to conduct source research.
- Do not summarize the law or treat this checklist as completed verification.
End with:
"Sources checked: [SOURCES].
Reviewed by: [ATTORNEY].
Date: [DATE].
Court's local rules on AI use checked: [YES / NO].
Unresolved issues: [ISSUES]."
Resolve outstanding issues before submission. Keep the completed checklist as a practical record of the review, rather than treating the generated checklist as evidence that verification occurred.
What does California’s 2026 update add?
California’s 2026 Practical Guidance addresses agentic AI and applies professional duties to its use. It replaces COPRAC’s original November 2023 guidance and addresses agentic AI at the California Supreme Court’s request.
The checklist above incorporates its requirements on strategic control, review, communication, fees, candor and supervision. For agentic workflows, the operational question is whether a lawyer meaningfully supervises substantive legal determinations and filing. The guidance prohibits deploying systems for those tasks without meaningful lawyer supervision.
For billing, preserve the distinction between actual hourly work and AI costs. Time spent crafting prompts and reviewing outputs may be charged; hourly billing must reflect actual time. Marking up AI costs requires informed written consent.
California is a state example, not a substitute for checking the rules where you practice.
What are this checklist’s limits?
This is general information, not legal advice or a guarantee of compliance. A licensed lawyer in the relevant jurisdiction must check how applicable professional rules and court requirements apply to the matter.
Local review should cover:
- The tool’s data handling and any required informed consent.
- Client communication about the proposed AI use.
- The court’s local rules on AI use.
- Flat fees, expense charges and consent for markups.
- Supervision and training for the planned workflow.
For how we assess tools, see the methodology. Tool testing cannot replace review of contractual terms or jurisdiction-specific professional duties.
FAQ
Must I tell every client that I use AI?
Opinion 512 distinguishes communication from informed consent. Rule 1.6 requires informed consent before representation information enters a self-learning tool. Rule 1.4 may require telling the client about GAI use even when that consent requirement does not apply. California requires lawyers to consider disclosure of intended AI use, how it will be used, and its benefits and risks.
Can I brainstorm without client consent?
Opinion 512 says idea generation that does not input information relating to the representation does not require informed client consent. Keep that condition in view: removing a client’s name does not necessarily remove representation information. Before entering such information, assess confidentiality risks and whether the self-learning-tool consent requirement applies.
Can I charge for time spent prompting?
California’s 2026 guidance permits charging for actual time spent crafting prompts and reviewing outputs; hourly billing must reflect actual time. Opinion 512 also warns that an unchanged flat fee may be unreasonable when GAI makes the task much faster. General office overhead may not be billed, and a surcharge on actual GAI expense requires disclosure.
Must I disclose AI use to the court?
Consult the court’s local rules on AI use, as Opinion 512’s footnote 47 directs. Separately, Rules 3.1 and 3.3 require review of outputs, including citations, before court submission. California’s guidance states that candor cannot be delegated to AI. A completed citation check does not answer the separate question of what the court’s rules require.
Sources
Links checked
- ABA Formal Opinion 512, Generative Artificial Intelligence Tools (July 29, 2024), checked 2026-10-20.
- State Bar of California, Practical Guidance for the Use of Generative AI in the Practice of Law (2026 update), checked 2026-10-20.
- State Bar of California, Artificial Intelligence resources, checked 2026-10-19.