Texas Opinion 705 on Generative AI, Explained for Small Firms

Texas has now told its lawyers, in plain terms, how the existing rules of professional conduct apply to generative AI. In February 2025, the Professional Ethics Committee for the State Bar of Texas issued Opinion 705, the state's first formal guidance on the ethical use of tools like ChatGPT, Claude, and the AI features built into legal research and drafting platforms.

The opinion is short and it does not announce new rules. It reads the duties already in the Texas Disciplinary Rules of Professional Conduct onto a new fact pattern. For lawyers and managing partners at firms of three to thirty attorneys, that is the useful part: nothing here requires a new committee or a six-figure software budget. It requires that a few familiar obligations be applied with care.

What Opinion 705 is, and what it is not

Opinion 705 answers a single question: what ethical issues are raised under the Texas Disciplinary Rules of Professional Conduct when a lawyer uses generative AI in the practice of law. The Committee's answer is that AI is permitted, that it can serve clients well, and that the responsibility for the work never leaves the lawyer.

The opinion grounds itself in rules that predate the technology by decades. It draws on Rule 1.01 (competent and diligent representation), Rule 1.05 (confidentiality of information), Rule 3.03 (candor toward the tribunal), and the related litigation-conduct rules, along with Rule 5.03, which governs a lawyer's responsibility for nonlawyer assistance. The framing matters. The Committee is not treating an AI tool as a novelty to be regulated separately. It is treating the tool the way it would treat a junior associate, a contract paralegal, or an outside vendor: a resource you may use, and whose output you remain answerable for.

Two practical consequences follow. First, Opinion 705 is advisory. It interprets the rules but does not itself carry disciplinary force, and it does not displace any standing order a particular court may impose on AI use in filings. Second, because it rests on existing duties, a lawyer who was already careful about supervision, confidentiality, and candor is most of the way to compliance. The opinion is best read as a checklist for habits you should already have.

Competence now plainly includes understanding the tool

The center of Opinion 705 is competence. Under Rule 1.01, a Texas lawyer must provide competent representation, and the Committee concludes that this duty extends to a working understanding of the technology a lawyer chooses to use. Before using a generative AI product for client work, a lawyer must understand, to a reasonable degree, how the tool functions, including its known limitations.

The standard is not that you become an engineer. It is that you understand enough to evaluate risk. A lawyer who does not grasp that a general-purpose chatbot predicts plausible text, rather than retrieving verified law, cannot sensibly judge when its output is safe to rely on. The Committee ties competence directly to this point: only a lawyer with a current and reasonable understanding of the technology can assess the risk of fabricated or inaccurate answers in a given task.

This has a few concrete implications for a small firm:

• The level of understanding scales with the stakes. Using AI to brainstorm deposition topics is lower risk than using it to draft a brief that will be filed.
• Competence is current, not one-time. Tools change, and a model's behavior this quarter may differ from last. The duty to understand the tool is ongoing.
• Supervisory lawyers carry this duty for their teams. Under Rule 5.03, a partner who lets associates or staff use AI is responsible for reasonable measures to ensure the conduct is compatible with the lawyer's own obligations.

Hallucinations and the duty to verify output

Opinion 705 is candid about the core failure mode of generative AI: it can produce confident, well-formatted, and entirely false content, including invented case citations. The opinion's response is the verification duty. A lawyer may not simply pass AI output through to a client, an adversary, or a court. The lawyer must independently confirm that what the tool produced is accurate before relying on it.

The reason this duty has teeth is that courts have already imposed real consequences. In Mata v. Avianca, Inc., decided in the U.S. District Court for the Southern District of New York in June 2023, Judge P. Kevin Castel sanctioned plaintiff's counsel under Federal Rule of Civil Procedure 11 after they filed a brief containing fictitious cases generated by ChatGPT, complete with fabricated quotations and citations. The court imposed a $5,000 penalty on the two attorneys and their firm and required them to notify the judges who had been falsely named as authors of the invented opinions. The lesson the opinion implicitly draws is direct: the tool's confidence is not evidence of its accuracy, and the lawyer who signs the filing owns the error.

Verification is also calibrated to the task. Confirming a citation means reading the actual opinion and checking that it exists, that it says what the draft claims, and that it remains good law. Confirming a summary of a contract means comparing it against the contract. The practical rule for a firm is simple to state: treat every factual or legal assertion produced by an AI tool as unverified until a lawyer has checked it against a primary source.

Read the terms before you input confidential information

The confidentiality analysis in Opinion 705 is the part small firms most often overlook, and it turns on the architecture of the tools themselves. Under Rule 1.05, a lawyer must protect confidential client information. The opinion warns that some generative AI systems retain the text users submit and may use it to train the model or surface it, in some form, in responses to later users. Inputting a client's confidential facts into such a system can therefore amount to a disclosure.

The Committee's instruction is to vet the tool before you feed it anything sensitive. The lawyer should be reasonably satisfied, before inputting confidential information, that the tool will not reveal that information to others or use it to the client's disadvantage. In ordinary practice that means reading the product's terms of service and data-handling policy and answering a short list of questions:

• Does the provider use submitted content to train its models, and can that be turned off?
• Is the data retained, and for how long, and who can access it?
• Is there an enterprise or business tier that contractually excludes your inputs from training and limits retention?
• Where is the data stored and processed, and does the provider's security posture match the sensitivity of the matter?

Many consumer-grade AND business offerings differ sharply on exactly these points, and the difference is usually a settings toggle or a contract tier rather than a different product. The defensible posture for a firm is to designate which tools are approved for confidential work, configure them to exclude training use where possible, and prohibit pasting client information into anything that has not been vetted. Where doubt remains, the conservative path is to anonymize the input or obtain the client's informed consent.

Bill for time worked, not time saved

Opinion 705 closes a tempting loophole. Generative AI can compress a task that once took hours into minutes. The opinion makes clear that a lawyer billing by the hour may not bill for time that was not actually spent. If AI drafts a first version of a memo in five minutes, the lawyer bills the time genuinely worked on it, including review and revision, not the time the task would have consumed without the tool. Charging a client for hours that efficiency eliminated would render the fee unreasonable.

The opinion treats the costs of the tool itself differently. Reasonable charges directly attributable to using a generative AI service, such as a per-matter usage fee, may be passed to the client, but only with the client's prior agreement and an appropriate disclosure. General overhead, by contrast, is the firm's to absorb. The clean way to handle this is to address AI in the engagement letter: state whether AI tools may be used, how time is billed, and whether any direct AI costs will be charged through.

This billing guidance aligns with ABA Formal Opinion 512, "Generative Artificial Intelligence Tools," issued by the ABA Standing Committee on Ethics and Professional Responsibility on July 29, 2024. That opinion reaches the same place on fees: a lawyer generally may not bill a client for the time spent learning to use an AI tool, and any cost passed through must be reasonable and disclosed. Reading the two together gives a small firm a stable, national-plus-Texas baseline rather than a single state's outlier view.

A short compliance checklist for a 3-to-30-lawyer firm

Opinion 705 does not demand a heavy program. A firm can meet its substance with a few written habits:

• Approve specific tools. Decide which AI products are cleared for confidential work, after reviewing their terms, and put the list in writing.
• Configure for confidentiality. Turn off training on your inputs where the product allows it, and prefer business or enterprise tiers for client matters.
• Verify before you file or send. Require that every citation and factual claim from an AI tool be checked against a primary source by a lawyer.
• Supervise the team. Make clear that associates and staff may use only approved tools in approved ways, consistent with Rule 5.03.
• Bill honestly. Charge for time worked, not time saved, and disclose any AI costs you intend to pass through.
• Update the engagement letter. Address AI use, verification, and billing so the client's expectations are set at the outset.
• Watch the courtroom rules. Confirm whether any judge or court before whom you appear has a standing order on AI use or disclosure in filings.

None of this is exotic. It is the ordinary discipline of competence, confidentiality, candor, supervision, and reasonable fees, applied to a tool that happens to be new. That is precisely how Opinion 705 frames Texas lawyer AI ethics, and it is why a careful firm can adopt these tools with confidence rather than anxiety: the duties have not changed, only the setting in which they are exercised.

This is general information for lawyers and law-firm leaders, not legal advice, and it does not create an attorney-client relationship. The authorities are cited so you can read them yourself.