When do lawyers have to tell clients they're using AI?

The question sounds simple. Most of the answers you will find online are not. They either say "always disclose" (overcautious, not what the rules require) or "no duty exists" (technically true in a narrow sense, but misleading enough to get you in trouble).

The honest answer is: it depends on how you used AI and what it touched. There is no blanket disclosure requirement for using AI in legal work. There are, however, at least four distinct rules that can each create a duty to tell your client, and most lawyers only know about one of them.

What ABA Opinion 512 actually says

ABA Formal Opinion 512, issued July 2024, is the most cited source on this question. What it actually says on disclosure is more nuanced than most summaries suggest.

The opinion imposes no blanket duty to tell clients you used AI. Running a clause through a drafting tool, summarising a deposition with AI-assisted software, or using a legal research platform with generative AI features: none of these automatically require a client conversation. The rules are not triggered by the use of AI itself.

What triggers them is what happens as a result. Opinion 512 identifies three specific situations where existing Model Rules kick in and create a disclosure obligation.

First, Model Rule 1.4 on communication. If you use AI to inform an important decision in the representation (something that materially affects strategy, scope, or outcome), the client needs enough information to make an informed decision. An AI-generated risk assessment that you simply relay to the client without exercising independent judgment is one example of where Rule 1.4 is in play.

Second, Model Rule 1.5 on fees. This is the trigger most lawyers underestimate. If AI saves you four hours on a matter and you bill as though it didn't (either at the original estimate or at the hours you would have spent without the tool), you have a fees problem, not just an AI problem. The Opinion is direct on this: the efficiency gains from AI must be reflected in what you charge.

Third, Model Rule 1.6 on confidentiality. If inputting client information into an AI tool creates a material risk of disclosure to third parties, the client needs to consent. Consumer-tier tools that retain and train on inputs are the obvious case. For a detailed breakdown of which tools retain what, see our confidentiality checklist.

The UK position: what the SRA actually requires

The SRA's February 2026 compliance guidance is more explicit on one point than any of the ABA materials: if a client is directly interacting with AI, that must be clear. An AI chatbot handling intake queries, an automated document portal, a tool that generates advice without a solicitor reading it first: clients need to know.

For background drafting assistance, the SRA has not taken the same position. A solicitor using an AI tool to draft a letter, check a contract, or summarise research, and then reviewing and signing off on the output, is not currently required to disclose that to the client as a separate obligation. The underlying duties apply: you remain personally responsible for every piece of work, and the output must be accurate.

What the SRA does say is that firms should consider whether to address AI use in the client care letter, based on the circumstances. That is not a mandate. It is a very clear signal that in any disciplinary situation, a firm without any mention of AI in its client care process will be in a harder position than one that addressed it upfront.

The SRA's existing rules on costs and billing apply directly. If AI reduces the time spent on a matter, clients are entitled to the benefit of that saving. Double-billing (charging for AI-generated work at manual rates without acknowledging the efficiency) creates a costs transparency problem under the SRA Transparency Rules.

The billing trigger: the one most lawyers miss

Florida Bar Opinion 24-1 made this concrete on the US side. It requires disclosure when AI use affects client billing or costs, not as a separate AI rule, but under the existing prohibition on unreasonable fees in Rule 4-1.5. If you charged the same fee with AI as you would have without it, on a matter where AI genuinely accelerated the work, that is worth examining.

The practical implication is that disclosure and billing are connected questions. A firm that wants to keep its fee structures private by not mentioning AI has made the calculation the wrong way round. The safer position is transparency about AI use and the associated efficiency, rather than leaving open the question of whether fees were inflated.

California's proposed amendments to its professional conduct rules, which entered public comment in May 2026, would codify this more formally: lawyers would be required to communicate sufficiently about AI use when it presents a significant risk or materially affects the scope, cost, or manner of representation. The proposals are not final, but the direction of travel is clear.

Court filings: a separate and stricter regime

Disclosure to clients and disclosure to courts are different obligations, and the court side is moving considerably faster.

As of mid-2026, more than 300 federal and state courts have some form of AI disclosure requirement for filings. The rules vary significantly. Some courts require certification that no AI was used in preparing substantive content. Others require disclosure of any AI involvement, with a certification that a human attorney verified the output. A handful have banned AI-assisted drafting for filings entirely.

The sanction record for 2026 should be enough to make this a standing item on any litigator's checklist. Courts imposed at least $145,000 in sanctions for AI-related filing failures in Q1 2026 alone. In one Florida case, Judge David Leibowitz found repeated and bad-faith AI misuse and ordered $86,000 in fees, then dismissed all four related federal cases. The Sixth Circuit, in Whiting v. City of Athens, imposed $15,000 per attorney plus full reimbursement of opposing counsel's fees and a disciplinary referral.

None of these cases turned on whether AI was disclosed. They turned on AI being used carelessly, with fabricated citations filed as though they were real. But courts that have AI disclosure requirements treat the failure to certify as a separate violation, regardless of whether the underlying content was accurate.

Check your court's standing orders before every filing. They change without announcement. The Tracelaw AI disclosure tracker is the most current resource for monitoring which courts have rules and what they require.

The engagement letter: the practical fix

Most disclosure questions disappear if you address AI use in the engagement letter. Not a lengthy policy document: a short paragraph that tells the client: your firm uses AI tools to assist with research and drafting; all output is reviewed and verified by a qualified lawyer; you do not share client-identifying information with consumer AI tools; and any efficiency savings are reflected in how you charge.

That paragraph does several things at once. It satisfies the Rule 1.4 communication obligation for most standard AI uses. It creates a record that clients consented to the general use of AI, which covers you on the Rule 1.6 question for tools with signed data processing agreements. And it removes any ambiguity about billing: if you've told the client how AI affects your process and pricing, there is no latent fees dispute.

ABA Opinion 512 specifically warns that boilerplate consent buried in standard terms is unlikely to satisfy the informed consent requirement for situations involving confidential data. The paragraph in your engagement letter needs to be readable and specific enough that a client actually understands what they're agreeing to.

For the underlying question of which tools are safe to use with client data in the first place, the client intake and AI safety guide covers the line between safe process work and riskier matter content. And if your firm doesn't yet have a written AI policy (which is the governance document the engagement letter sits inside), the AI policy template is a one-afternoon fix.

The disclosure question will keep evolving. California's proposed rules are moving. The SRA has signalled a Good Practice Note on AI and client data is coming later in 2026. Courts will keep adding requirements. What won't change is that a short, honest paragraph in your engagement letter covers far more ground than the current rules strictly require, and costs you nothing to include.

Download the free guide at The Lawyer's AI Starter Kit for a ready-to-use engagement letter paragraph and a checklist of the specific triggers that create disclosure obligations under ABA Opinion 512 and SRA guidance. The best AI tools for lawyers page covers which tools have the DPA and zero-retention status needed for safe client-data use.

FAQ