Ethics and Compliance

ABA Formal Opinion 512 on Generative AI

The duties ABA Formal Opinion 512 applies to generative AI, from competence and confidentiality to fees, and the daily controls that put each in place.

AssistCall research teamPublished Reviewed by Faruque Amin, VP of Sales and Marketing6 min read

On this page
  1. What the opinion is, and is not
  2. Competence: know the tool, check the output
  3. Confidentiality: self-learning tools need informed consent
  4. Communication: when to tell clients
  5. Candor and meritorious claims
  6. Supervision: policies, training and oversight
  7. AI you did not choose: features inside existing software
  8. Fees: actual time only
  9. Client-facing tools at intake
  10. Example: rolling out a drafting tool
  11. A thirty-day plan
  12. Frequently asked questions
  13. References
Ethics and Compliance: ABA Formal Opinion 512 on Generative AIArticle coverETHICS AND COMPLIANCEABA Formal Opinion 512on Generative AIAva for LawyersInsightsEthics and Compliance: ABA Formal Opinion 512 on Generative AIArticle coverETHICS AND COMPLIANCEABA Formal Opinion512 on Generative AIAva for Lawyers

Key takeaways

  • Opinion 512, issued July 29, 2024, applies existing rules to generative AI: competence, confidentiality, communication, candor, supervision and fees[1].
  • Lawyers need a reasonable understanding of a specific tool's capabilities and limits and must independently verify what it produces[1].
  • Informed consent is needed before client information goes into a self-learning tool, and general boilerplate in an engagement letter is not enough[2].
  • Bill only time actually spent; do not bill clients for learning a tool[1][2].

What the opinion is, and is not

On July 29, 2024, the ABA's Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512, its guidance on lawyers' use of generative AI tools[1]. It creates no new rule. It explains how the Model Rules lawyers already follow apply to tools that draft, summarize and answer questions. ABA opinions are persuasive rather than binding, and states issue their own guidance, but 512 has become the common reference point for firm AI policies.

Adoption makes the opinion practical rather than theoretical. In Thomson Reuters Institute's 2025 survey, 26% of legal professionals said they already used generative AI, up from 14% the year before[3].

Legal professionals already using generative AI
Legal professionals already using generative AIColumn chart. 2024 report: 14%; 2025 report: 26%.0%10%20%30%2024 report: 14%14%2024 report2025 report: 26%26%2025 reportLegal professionals already using generative AIColumn chart. 2024 report: 14%; 2025 report: 26%.0%10%20%30%2024 report: 14%14%2024 report2025 report: 26%26%2025 report

Source: Thomson Reuters Institute, 2025[3]

Show the data
ItemValue
2024 report14%
2025 report26%

Competence: know the tool, check the output

The opinion expects lawyers to have a reasonable understanding of the capabilities and limitations of the specific generative AI tool they use, and because these tools are only as good as their data, to independently verify the information they produce[1]. Verification is not optional even with tools built for lawyers: Stanford researchers found legal research products from major providers produced incorrect information more than 17% of the time in their tests, and one hallucinated more than 34% of the time[4].

  • Keep a short register of approved tools with what each is good at and known weaknesses.
  • Require a human check of every citation, quotation and factual statement before anything leaves the firm.
  • Scale review to risk: a client email needs a read; a brief needs every authority pulled and read.

The opinion's confidentiality analysis turns on whether a tool learns from what users put in. If information relating to the representation could be disclosed to people outside the firm through a self-learning tool, the client's informed consent is required first, and general boilerplate provisions in an engagement letter do not satisfy that requirement[2]. Consent has to be informed: the client needs to understand the risk, not just see the word AI in a long list.

  1. Classify each tool

    Read the vendor's terms and data settings. Record whether inputs are used for training, how long they are kept and who can access them.

  2. Default to non-learning

    Prefer tools or settings where inputs are not used to train models available to others.

  3. Get specific consent when needed

    If a matter requires a self-learning tool, explain the risk in plain terms and obtain the client's informed consent, separate from boilerplate.

  4. Minimize inputs

    Remove names and identifying details where the task does not need them.

Communication: when to tell clients

Not every use must be disclosed, but the duty to communicate can require it. Summaries of the opinion note that disclosure may be in the client's interest, especially where AI use affects the fee or significantly influences a decision in the representation[2]. Clients often do not know: 71% of law firm clients surveyed by Thomson Reuters did not know whether their firms used generative AI tools[5].

A practical approach is a short, honest paragraph in the engagement letter describing how the firm uses AI tools and how output is reviewed, plus a direct conversation in any matter where AI use will be significant or where the client has asked about it. Our article on disclosing AI use to clients offers wording.

Candor and meritorious claims

Rules on meritorious claims, candor to the tribunal and dishonesty apply to AI-assisted filings exactly as they do to any other. A lawyer who files a brief citing authority that does not exist has the same problem whether the citation came from a junior associate or a tool. Courts continue to sanction lawyers for fabricated citations, and some require certifications about AI use in filings. Check local rules and standing orders in every court where you appear.

Supervision: policies, training and oversight

Managerial and supervisory lawyers must make reasonable efforts to ensure that lawyers, staff and vendors use AI consistently with the rules. Commentary on the opinion describes this as a governance program: written policies, education on those policies and supervision of how tools are used[2]. Training is the weak point in most firms. Thomson Reuters found that while 89% of professionals see use cases for generative AI, 64% had received no training on using it for their work[6].

The duties Formal Opinion 512 discusses and a control that puts each into practice.
DutyRuleDaily control
Competence1.1Approved-tool list and a verification step for all output
Confidentiality1.6Tool classification, data settings and informed consent where needed
Communication1.4Engagement letter paragraph and matter-level conversations
Candor and merit3.1, 3.3, 8.4(c)Citation checks before filing; court rule review
Supervision5.1, 5.3Written policy, training log, periodic review
Fees1.5Bill actual time; disclose any tool costs passed on

AI you did not choose: features inside existing software

Many firms meet generative AI first as a new button in software they already use: email that drafts replies, a document system that summarizes files, a video platform that writes meeting notes. The opinion's duties apply to those features just as much as to a tool the firm bought on purpose, and they are easier to miss because nobody approved them.

  • Ask each software vendor which AI features are switched on by default and what happens to the data they process.
  • Turn off features the firm has not reviewed, especially meeting assistants that record and summarize client calls.
  • Add newly released features to the approved-tool review, rather than letting them arrive through an update.
  • Tell staff that the policy covers built-in features, not only standalone AI tools.

Fees: actual time only

Fees must be reasonable, and when AI makes work faster, firms must take that into account. As ABA Journal coverage put it, it is not permissible to charge a client for more hours than were actually spent: if a tool turns hours of document searching into minutes, the firm bills the minutes[1]. A lawyer also cannot bill a client for learning how to use a tool, and only extraordinary costs associated with a tool may be charged to the client, with the client's knowledge and consent[2].

Flat fees are not a loophole. A flat fee must still be reasonable, and a fee set when a task took ten hours may become unreasonable when it takes one.

Client-facing tools at intake

The opinion's duties apply with particular force to tools that talk to the public, because the person on the other end may share confidential facts before anyone at the firm is involved. Supervision means the firm approves the script, decides what the tool may and may not say, and reviews conversations. If an SI Employee such as Ava answers your calls, those approvals come first: the firm signs off the intake questions, the AI and recording disclosures and the rule that it never gives legal advice.

Example: rolling out a drafting tool

Example: a fictional eight-lawyer firm wants to use a drafting assistant for demand letters. The managing partner reads the vendor terms and confirms inputs are not used for training and are deleted after a set period. The firm adds the tool to its approved list, writes a one-page procedure requiring a lawyer to check every fact and figure, trains staff in a one-hour session and logs attendance. Its engagement letter gains a paragraph explaining that the firm uses AI tools under lawyer review and bills only actual time. Billing entries for AI-assisted drafts record the real review time.

A thirty-day plan

  1. Week 1: list every AI tool and AI feature in existing software that anyone in the firm uses.
  2. Week 2: classify each for data use and decide approved, restricted or banned.
  3. Week 3: adopt a written policy, update the engagement letter and set billing guidance.
  4. Week 4: train everyone, record attendance and schedule a review in six months.

For how states have built on the opinion, see California's guidance and Florida Bar Opinion 24-1.

Frequently asked questions

What is ABA Formal Opinion 512?

The ABA ethics committee's July 2024 opinion on generative AI. It applies the existing Model Rules on competence, confidentiality, communication, candor, supervision and fees to AI tools.

Does Opinion 512 require client consent to use AI?

Informed consent is required before information relating to the representation is put into a self-learning tool that could disclose it outside the firm. General boilerplate in an engagement letter is not sufficient.

Can lawyers bill for time saved by AI?

No. Hourly fees must reflect time actually spent, and lawyers should not bill clients for learning a tool. Flat fees must remain reasonable.

Is Opinion 512 binding?

ABA formal opinions are persuasive, not binding. State rules and state bar opinions govern, and many states have issued their own guidance consistent with 512.

References

  1. ABA Journal (2024). Take heed before using artificial intelligence, new ABA ethics opinion says (opens in a new tab) โ†‘

  2. Robinson+Cole Data Privacy + Security Insider, via JD Supra (2024). American Bar Association Issues Formal Opinion on Use of Generative AI Tools (opens in a new tab) โ†‘

  3. Thomson Reuters Institute (2025). 2025 GenAI report: Executive summary for legal professionals (opens in a new tab) โ†‘

  4. Stanford Institute for Human-Centered Artificial Intelligence (HAI) (2024). AI on Trial: Legal Models Hallucinate in 1 out of 6 (or More) Benchmarking Queries (opens in a new tab) โ†‘

  5. Thomson Reuters Institute (2025). 2025 GenAI report: Executive summary for legal professionals (opens in a new tab) โ†‘

  6. Thomson Reuters (2025). Generative AI Adoption Nearly Doubles as Professional Services Reach Crossroads (opens in a new tab) โ†‘

Book a demo

See Ava handle your calls.

Book a short demo. Weโ€™ll walk through your calls and workflow, and agree what Ava should take care of.

Book a demo