Key takeaways

  • Both opinions apply existing rules of professional conduct to AI. They create no new duties, and neither forbids generative AI.
  • Competence means understanding what a tool can and cannot do and checking every output. A lawyer cannot replace their judgment with the tool's.
  • Confidentiality is the hard part. Read the tool's terms, find out whether it learns from your inputs, and get informed consent before client information goes to a third-party tool. Boilerplate in an engagement letter may not be enough.
  • On fees, both opinions say you cannot bill for time spent learning a tool, cannot double-bill, and cannot bill more than the time actually spent or an unreasonable flat fee. Florida treats general AI subscription costs as overhead.
  • Firms need a written AI policy, training and a list of approved tools. A tool that runs inside your own environment can remove most of the confidentiality risk.

Two opinions, one message

Generative AI tools draft memos, summarize depositions, review contracts and answer client questions. Lawyers want the time savings. Bar regulators want to know that the profession's duties still hold. Two ethics opinions give the best guidance so far for Florida lawyers: Florida Bar Ethics Opinion 24-1, issued on January 19, 2024, and the American Bar Association's Formal Opinion 512, "Generative Artificial Intelligence Tools," issued on July 29, 2024.

Both are advisory. Opinion 512 applies the ABA Model Rules, which become binding only where a state adopts them, and Florida's opinion is guidance from the state bar. Both reach the same conclusion: a lawyer may use generative AI, but only to the extent the lawyer can meet the existing ethical duties. This article summarizes them in plain language. It is not legal advice. Check the opinions themselves and your own rules.

A note on sources: we read the full text of Opinion 512 (15 pages) in the copy the ABA published, and checked this article against it. The ABA's own pages block automated access, so the links below point to Internet Archive copies of the ABA documents. We also read the Florida Bar's opinion in full. Summaries of Opinion 512 from the National Law Review and the UNC Law Library are listed for context. Read the opinion yourself before you adopt a policy.

Competence: know the tool, check the answer

Model Rule 1.1 requires competent representation, and its comment 8 says a lawyer should keep abreast of changes in the law and its practice, "including the benefits and risks associated with relevant technology." Under Opinion 512 a lawyer must reasonably understand the capabilities and limits of the technology, stay alert to its risks, and keep up as the tools change. The central risk is the tool's tendency to produce confident errors, often called hallucinations. Lawyers must independently verify output, and the opinion says a lawyer may not abdicate professional judgment to the tool. The depth of review depends on the tool and the task: a lawyer summarizing lengthy contracts need not necessarily re-read every document if the tool has proven reliable on that task.

Florida's opinion says the same from a different angle. Lawyers must verify the accuracy and sufficiency of AI-generated research, and a failure to do so can violate duties of competence, of avoiding frivolous claims, of candor to a tribunal and of truthfulness to others. The opinion adds that a lawyer may not delegate tasks that constitute the practice of law or require professional judgment.

Opinion 512 connects this to candor and meritorious claims, Rules 3.3 and 3.1. A hallucinated citation in a filing is the standard example. The opinion's footnote tells lawyers to consult the applicable court's local rules on AI use.

Confidentiality: the central risk

Rule 1.6(c) requires a lawyer to "make reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information relating to the representation of a client." Opinion 512 asks lawyers to think about two kinds of exposure before they type anything into a tool. The first is outside disclosure to the vendor and anyone who can reach its systems. The second is inside disclosure: information entered into a firm-wide tool may be visible to other lawyers at the firm and could breach an ethical wall or an access control. Some tools also use inputs for self-learning, which could surface client information to other users.

The practical steps the opinions describe are the same for any vendor.

  • Read the privacy policy, terms of use and contract. Find out what the vendor retains, who can read it, whether it trains models on your inputs, and where the data is stored. Florida's opinion tells lawyers to research data retention, data sharing and self-learning policies.
  • Consult your IT or security professionals where the technical answer is unclear.
  • Obtain informed consent when client information will go to a third-party tool. Opinion 512 says the client must be told of the use and its risks, and the opinion says that merely adding general, boiler-plate provisions to engagement letters purporting to authorize the use of AI is not sufficient. Florida recommends informed consent for third-party tools and notes that an in-house tool that does not send data to a third party may not require it.
  • Limit what goes in. Strip names and identifiers where the task does not need them, and keep matter data out of general-purpose consumer chat tools.

This is where the choice of tool matters most. A consumer chatbot and a model that runs inside your own environment raise very different confidentiality questions. Our article on private AI for sensitive data and the one on local open-weight models versus frontier APIs describe the options, including running a model on your own hardware or in a private cloud tenant.

Communication with clients

Rule 1.4 requires keeping clients reasonably informed. Opinion 512 says disclosure of AI use may be in the client's best interest, especially when the tool affects the fee or when its output will significantly influence a decision in the matter. The disclosure can go in the engagement letter, and direct communication beforehand may be appropriate. If a client asks whether AI is used on their matter, the lawyer must answer honestly.

Fees and costs

Both opinions address billing directly, and their message is consistent. Fees must be reasonable under Rule 1.5, and AI does not change that.

  • No billing for learning. A lawyer cannot bill a client for time spent learning how to use a tool for general practice. Florida adds that lawyers should not bill for time spent learning basic AI competence.
  • No double-billing. Florida states that use of generative AI does not permit improper billing practices such as double-billing. Under an hourly arrangement, bill for the time actually spent, not the time the tool saved.
  • Efficiency is not a windfall. An efficient hourly lawyer still may charge only for the time actually expended, and a tool that makes the work much faster can make an unchanged flat fee unreasonable. Flat fees are not exempt: the opinion says a fee for which little or no work was performed is unreasonable.
  • Costs. A lawyer may bill the actual cost of AI used on a client's matter as an expense when the client has been told in advance, preferably in writing, and should pass along any discounts. A tool that works like equipping and maintaining a practice, such as a grammar checker inside word-processing software, is overhead and is not billed absent a contrary advance disclosure. Florida also treats general subscription costs as overhead when the per-client cost cannot be determined.

Supervision and governance

Rules 5.1 and 5.3 make partners and managers responsible for the conduct of lawyers and nonlawyers under them. Opinion 512 applies those rules to AI. Managers should adopt clear policies on which tools are allowed and what data may be entered, train lawyers and staff on the risks, and supervise use. Opinion 512 also reminds firms that work outsourced to third parties that uses AI must meet the same ethical obligations. Florida says that lawyers must supervise generative AI much as they supervise nonlawyer assistants and remain responsible for the work product.

Client-facing chatbots and advertising

Florida's opinion adds a point that firms with websites should note. A chatbot used for intake or communication must follow the lawyer advertising rules. It must tell the user that they are communicating with an AI program and not with a lawyer or firm employee. The opinion also cautions against bots that give legal advice, suggests screening out people who already have counsel, and allows a lawyer to advertise use of generative AI but not to claim that the lawyer's AI is superior to others' unless the claim can be objectively established.

A practical AI policy for a law firm

A one- to two-page policy is enough to start. It should cover the following.

  • Approved tools. A named list, with the data classes each may receive. Everything else is off limits for client matters.
  • Data rules. What may never be entered, such as sealed, privileged or regulated data, unless the tool runs in a controlled environment.
  • Verification. Every citation, quotation and factual statement from a tool is checked against a primary source before it leaves the firm.
  • Disclosure and consent. Language for the engagement letter, a conversation script and a rule for matters where the client asks you not to use AI.
  • Billing. How time is recorded when AI helps, and how any tool cost is treated.
  • Training. Initial and refresher sessions, with records.
  • Incident handling. What to do if confidential information is entered by mistake or a tool vendor reports a breach.
  • Review. A named person owns the policy and revisits it at least twice a year, because the tools and the rules are changing.

Choosing the tool: questions for a vendor

  • Does the vendor train on our inputs or outputs? Can that be contractually excluded?
  • Where is the data processed and stored, and how long is it kept? Can we delete it?
  • Who at the vendor can see our data, and is access logged?
  • Is the tool segmented by matter, so that one lawyer's input is not visible to another who is walled off?
  • Does it integrate with our identity system and enforce MFA and role-based access?
  • What happens on a breach, and what does the contract say about notice and liability?
  • Can the tool show the sources behind its answers so that a lawyer can verify them quickly?

How Clayton Cybernetics can help

Our Super Intelligence practice builds private AI for organizations that cannot send client data to a public service. That includes self-hosted models, retrieval over a firm's own documents with matter-level permissions, logging for supervision, and integration with existing identity systems. Our team has built a custom, self-hosted practice platform for a law firm, with AI in every layer. We can review the AI tools your lawyers already use, draft a policy that fits Opinion 512 and Florida Opinion 24-1, and design a private alternative for the work that is too sensitive for a public tool. Call (305) 404-9404 to start.

Sources

This article is general information, not legal advice or ethics advice. It was checked against the text of ABA Formal Opinion 512 as published on July 29, 2024; later ABA materials may differ. Rules differ by state and court. Confirm with the opinion itself, your bar's ethics counsel and any court's standing orders. Last reviewed .

Keep reading

Need help with this?

American‑made Super Intelligence: AI your team can learn, trust and deploy.

Super Intelligence