New York Bar Tells Lawyers: Bill for Checking AI, Not for the Time It Saved

On October 31, 2026, the New York State Bar Association’s House of Delegates adopted “Legal Ethics Guidelines for Lawyers’ Use of Artificial Intelligence,” a report from the Committee on Artificial Intelligence and Emerging Technologies. The document is guidance for New York lawyers, not an amendment to the New York Rules of Professional Conduct. It still matters, because it tells firms how bar leaders expect existing duties—competence, confidentiality, supervision, fees, and candor to tribunals—to apply when generative and agentic tools sit inside the matter file.

What the House of Delegates adopted

According to the association’s announcement, president Taa Grays framed the report as a resource for responsible use rather than a ban. “Generative AI is indelibly changing the practice of law,” she said. “As part of our mission to be the leading voice for the legal profession, we are providing clear guidelines on the use of AI tools as a resource so attorneys can use AI responsibly and ethically in their practices.” She flagged two practical perils the guidelines are meant to reduce: inadvertently practicing in a jurisdiction where the lawyer is not licensed, and failing to notify clients or courts when AI use requires notice.

Vivian Wesson, executive vice president, general counsel, and chief compliance officer of The Board of Pensions of the Presbyterian Church (U.S.A.), chairs the committee that wrote the report. Her public comment stressed continuity of ethical principles while technology changes, and the committee’s aim of helping lawyers use tools without abandoning duties to clients, courts, and the administration of justice.

The NYSBA release and subsequent coverage in the ABA Journal are the sources for the points below. The guidelines extrapolate from the New York Rules of Professional Conduct; they do not rewrite those rules, and they are not a court order.

Billing: charge verification, not the shortcut

The fee advice is the part most firms will feel first. Attorneys may charge for time spent verifying an AI tool’s output. They should not bill for time the tool saved, and they should not bill clients for time spent learning how to use the tool. Fees must remain reasonable, and specific charges should be disclosed. Buying and maintaining AI software is treated as overhead. That cost generally may not be passed through as a disbursement unless the expense was specially incurred for a particular matter and the basis for the charge is disclosed to the client.

Reporting summarized by the ABA Journal, drawing on Law360, adds a related distinction: lawyers generally do not have to tell clients about AI used for routine administrative tasks—word processing, spelling and grammar checks, calendaring, or timekeeping—when that use does not affect fees. The disclosure duty rises when AI changes the economics of the engagement or when court or client rules demand notice.

That billing line sits next to a broader competence problem. California has already moved from soft guidance toward statute on verification and court disclosure; firms comparing state approaches should read how California’s SB 574 requires lawyers to verify AI output and keep client secrets out of open models. New York’s report is bar guidance. California’s statute is law. Both point at the same habit: if a machine drafted it, a human still owns it.

Supervision, vendors, and the “non-lawyer” that is a model

The New York Rules do not speak in AI-native vocabulary, but the guidelines extrapolate a familiar result: lawyers remain responsible for the work product of non-lawyers they supervise, including when the “non-lawyer” is an AI system or a vendor’s AI pipeline. Approving a firm-wide tool or policy does not remove the duty to review work on each matter. Supervising attorneys should not assume that procurement sign-off equals matter-level quality control.

That duty reaches outside the firm. When a discovery vendor, contract-review platform, or freelance researcher uses AI on client data, the New York lawyer who engaged them still has to understand how the tool is configured, what data it retains, and whether its outputs were checked. The privacy risk is not theoretical. Law-firm AI platforms concentrate client information in new ways; the operational questions mapped in coverage of legal-engineering and client-data governance inside law firms are the same questions ethics counsel will ask after a leak or a hallucinated citation.

Conflicts, confidentiality, and agentic tools

The report flags conflicts of interest when a tool uses data that pertains to a party in a case—especially if training or retrieval mixes one client’s information into another matter. Confidentiality analysis under the New York Rules still controls. If an AI system learns from Client A in a way that can surface in Client B’s work, counsel needs a conflicts and confidentiality answer before the tool stays on.

Other points in the association’s summary:

• Check that results do not reflect biased training data, design, or deployment, particularly where outputs could affect rights, access, opportunity, or treatment.
• Keep legal decisions in human hands; do not let a model become the decision-maker of record.
• Watch unauthorized practice of law risks, including tools that deliver automated legal services or that present as a human to prospective clients across jurisdiction lines.
• Notify parties when AI is used for recording and summaries, as the report directs.
• Disclose AI use in court submissions where local rules already require it.
• Review the output of agentic AI systems—tools designed to complete tasks or goals with limited human input—before relying on them.

None of these points invents a new professional duty. They map existing duties onto tools that write, retrieve, summarize, and, increasingly, act. National and state AI governance debates continue in parallel; practitioners who want the wider legislative map can start with the U.S. state artificial intelligence governance laws overview and with proposals such as Connecticut’s proposed artificial intelligence act, which show how quickly bar guidance and statute can converge on the same failure modes.

A six-step firm checklist after the NYSBA vote

Guidance only helps if someone turns it into intake, engagement letters, and review habits.

1. Inventory AI tools by matter risk. Separate administrative helpers (grammar, calendaring) from tools that draft advice, analyze evidence, or talk to clients. Note which ones retain prompts or train on inputs.
2. Rewrite fee language. State that verification time is billable, learning time is not, and efficiency gains are not billed as phantom hours. Describe when an AI-related expense may be charged as a matter-specific disbursement.
3. Set a human-sign-off rule. Require a licensed attorney to approve any AI-assisted filing, opinion, or client-facing advice before it leaves the firm.
4. Extend supervision to vendors. Add AI-use, retention, and no-training clauses to vendor agreements; sample outputs on live matters; keep an audit trail of who reviewed what.
5. Run a conflicts screen for shared models. Ask whether the tool isolates client workspaces. If it cannot, treat multi-client use as a confidentiality and conflicts problem, not an IT preference.
6. Track court standing orders. Maintain a jurisdiction list of AI-disclosure rules for filings, and default to disclosure where the local rule is unclear and the use was substantive.

Federal and state AI safety bills will keep moving; the association’s report sits beside statutes such as the Artificial Intelligence Safety Measures Act debate rather than replacing them. Ethics guidance answers how lawyers bill and supervise. Legislatures answer what products may do. Firms need both files.

Frequently asked questions

Do the NYSBA guidelines change the New York Rules of Professional Conduct?

No. They are association guidance that interprets how existing rules apply to AI use. Courts and disciplinary authorities still look first to the Rules and to case-specific facts.

Can a New York lawyer bill a client for time saved by AI?

The guidelines say no. Billable time is for work performed, including verification of AI output. Time saved by the tool, and time spent learning the tool, should not be charged.

Is buying an AI subscription a client disbursement?

Generally no. The report treats acquisition and maintenance as overhead, unless the cost is specially incurred for a specific matter and disclosed.

Must every use of AI be disclosed to the client or the court?

Not necessarily for pure administrative tasks that do not affect fees. Substantive use, fee effects, client instructions, and local court rules can all require notice. When a standing order demands disclosure of AI-assisted filings, follow the order.

Who remains responsible if a vendor’s AI hallucinates?

The supervising lawyer. Firm approval of a vendor does not erase matter-level review duties.

The signature line is still human

Bar associations write reports when members are already using the tools and already making quiet billing and supervision choices that will look obvious in hindsight. New York’s October 31 vote does not ban generative AI. It says the efficiencies are real, the ethical shortcuts are not, and the person whose name is on the brief still has to read what the model produced. Firms that treat the guidelines as optional color will meet them again—in a fee dispute, a malpractice claim, or a court that asks who checked the citations.