The lawyer-specific case for per-matter AI disclosure
Lawyers occupy an unusual disclosure-risk position. The profession already runs on a denser web of duties of candor, competence, and communication than almost any other field, and the standing orders, state-bar opinions, and ethical-rule interpretations that have appeared since the Mata v. Avianca sanctions order in 2023 have layered AI-specific obligations on top of all of them. A solo practitioner in Chicago drafting a motion for a Northern District of Texas case has to think about three different disclosure layers at once: the court's standing order on the case, the home-state bar's opinion on AI competence, and the firm's own engagement letter with the client. By 2026, ignoring any one of them carries real downside.
The cheapest insurance is a per-matter disclosure block that satisfies the strictest layer that applies and is therefore safe in every layer below it. This generator emits exactly that — four small, copy-paste blocks shaped for the four moments where the disclosure must actually appear: under the signature block of a filing, in the engagement letter or status email, in the footer of any marketing-page or blog post the firm publishes, and at the start of any AI chatbot conversation on the firm website.
The four AI usages every law firm actually has
Not every AI touch needs the same disclosure, and overdisclosing to the wrong audience dilutes the signal. The generator's four presets map to the four cases that genuinely require disclosure under 2026 rules:
- Court filings drafted or research-assisted with AI. The single highest-stakes case. The court-order preset emits a four-element certification: AI used yes or no, name of the model, attorney-verified citations, and counsel signature. Long-form narrative version available for judges with standing orders that ask for more detail.
- Client communications where AI use is material. ABA Formal Opinion 512 does not require disclosure on every email, but it does require it when the AI use materially affects representation — an AI-drafted demand letter, an AI-summarized deposition transcript, an AI-built case-strategy memo. The preset emits a one-paragraph notice for the engagement letter and a per-matter line for status updates.
- Marketing copy and firm-website content. AI-drafted partner posts attributed to a named attorney, AI-generated headshots, AI-summarized practice-area pages, AI client testimonials — each touches Rule 7.1. The marketing preset emits a footer disclosure, an image label, and a flagged warning for jurisdictions (FL, NY, NJ) with stricter advertising rules.
- Client-intake chatbots. The Article 50, Rule 7.3, and SB 1001 triple-overlay. The chatbot preset emits the opening banner ("You are talking to an automated assistant, not a lawyer"), the first-message text, and the no-attorney-client-relationship-formed footer the engagement-rule cases require.
Court standing orders: the 2026 patchwork and what survives in every version
There is still no single federal rule on AI in court filings. The result is a patchwork: the Fifth Circuit considered and shelved a circuit-wide rule, the Northern District of Texas's Judge Starr order is the most-copied template, the Eastern District of Pennsylvania's order is among the most demanding, and a handful of state courts (notably in Illinois and New York) have published their own variants. The wording differs, but every version that has actually been enforced asks the same four things:
- Was generative AI used at all in drafting the filing.
- If yes, which system — named, with model version when applicable.
- Was every citation verified by a human attorney admitted in the jurisdiction, by reading the cited source.
- Signature of counsel of record acknowledging the certification under the same rules that govern Rule 11.
The generator's default court-filing preset emits exactly those four elements. The optional long-form preset adds three more fields some judges request: the role the AI played (drafting, research, summarization), whether the client was informed of AI use, and the lawyer's independent professional judgment that the resulting filing is non-frivolous. The American Bar Association tracks the standing-order list at the ABA, and the Law360 Generative AI tracker is the most current public list.
ABA Formal Opinion 512 in one paragraph (and what it actually changes)
ABA Formal Opinion 512 (July 2024) does not create new duties. It clarifies that the existing duties of competence (1.1), confidentiality (1.6), candor to the tribunal (3.3), communication (1.4), and reasonable fees (1.5) all apply to a lawyer's use of generative AI tools.
The practical consequence is that AI competence is now part of the duty of competence: a lawyer needs to understand the AI tool well enough to verify its output, must not transmit client-confidential information to a tool that does not provide adequate confidentiality protection, must inform the client when AI use is material to representation or where it might affect fees, and may not bill the client at attorney rates for time the AI did. Opinion 512 is permissive about routine AI use that does not affect the substance of representation — an in-firm research assistant that lawyers cross-check is fine without per-matter disclosure — and stricter the closer AI gets to the client's decisions. The generator's "ABA 512" preset emits the conservative version of the disclosure: a paragraph for the engagement letter that flags AI use is contemplated and a per-matter line for status emails on matters where AI use is material.
State-bar opinions: what to add on top of ABA 512
State bars have layered on top of 512 in ways worth a per-state preset:
- Florida Bar Ethics Opinion 24-1. Requires lawyers to obtain client consent before using a generative AI tool, requires confidentiality assessment of any third-party AI vendor, and flags non-lawyer-supervision concerns when AI is used by paralegals.
- California State Bar (November 2023 Practical Guidance). Adds a no-double-billing rule (you cannot bill the client for AI time at attorney rates) and emphasizes the duty of supervision over AI outputs.
- New York City Bar Opinion 2024-5. Treats AI tools as effectively non-lawyer assistants and requires they be supervised under Rule 5.3, with adequate confidentiality safeguards before any client information is shared.
- D.C. Bar Opinion 388. Aligns broadly with ABA 512 but is specific about prompt-injection and adversarial-input risk in litigation contexts.
The generator emits state-specific addenda you append to the ABA 512 block.
Marketing rules: where AI content crosses Rule 7.1
Rule 7.1's prohibition on false or misleading communications about a lawyer's services is the umbrella that catches almost every AI marketing failure mode. The three patterns that have produced actual disciplinary actions or sharp warnings since 2024:
- Synthetic attorney headshots. AI-generated photos of an "attorney" who does not exist, used in a small-firm site to suggest a deeper bench. Flatly misleading under Rule 7.1 and increasingly under state-level deepfake statutes.
- AI-authored partner posts. Long-form blog content under a named partner's byline that the partner did not write or substantively review. Misleading under Rule 7.1; possible Rule 8.4(c) issue (conduct involving dishonesty).
- AI-generated client testimonials. Generated reviews on the firm site or syndicated to legal directories. Categorically prohibited.
The marketing preset emits a footer ("This article was drafted with AI assistance and substantively reviewed by [attorney name], [bar number]"), an image label for AI illustrations ("AI illustration; not a photograph of firm personnel"), and a state-tier flag when the firm is admitted in jurisdictions with stricter advertising rules. For broader 7.1 background, the ABA Center for Professional Responsibility is the canonical source.
Intake chatbots: the triple overlay nobody catches the first time
A firm website chatbot looks like the easiest disclosure case and is, in fact, the most-frequently-missed. Three rules apply simultaneously:
- Article 50, EU AI Act — any person interacting with an AI system must be told. Binding August 2, 2026, applies the moment the chatbot is reachable from an EU IP.
- Rule 7.3 (analogue in every U.S. jurisdiction) — no live solicitation. An AI chatbot is treated as "live" when it replies in real time, so it must clearly identify as automated and must not initiate the conversation unsolicited.
- California SB 1001 and its New York/Illinois cousins — any bot that materially influences a commercial transaction must disclose its bot status.
The chatbot preset emits a single banner that satisfies all three plus a no-attorney-client-relationship-is-formed footer for the engagement-rule cases. It is the smallest preset in the generator and the one with the highest cost of getting wrong.
What this is not: legal advice
This is a free disclosure-label generator. It does not replace your malpractice carrier's guidance, your state bar's ethics hotline, or independent counsel on the specific facts of your matter.
Use it as a starting block. Have a partner or your firm's general counsel sign off on the engagement-letter and chatbot variants before they go live. Treat the court-filing certification as a template, not a substitute for reading the standing order on the case in front of you.
Compliance vs. theatre: what bad law-firm AI disclosure looks like
| Pattern | What it does | Status |
|---|---|---|
| One footer on the firm-wide "Legal" page | Reader of the article never sees it; court never sees it | Non-compliant |
| AI-drafted brief with hallucinated citations | Rule 11 sanctions, the Mata v. Avianca fact-pattern | Aggravated risk |
| Generic "some content may use AI" site-wide | Doesn't identify the matter, the model, or the verification | Borderline |
| Chatbot says "Welcome!" with no bot label | Article 50, Rule 7.3, and SB 1001 violations simultaneously | High risk |
| Per-matter four-part certification + ABA 512 letter + marketing footer + chatbot banner | Court, client, regulator, and bar all see the right disclosure | Best practice |
Workflow for a typical small-firm or solo practice
The pattern that actually scales: open the generator once at the firm level, build the standard variants — "court-filing default," "court-filing long form," "engagement-letter paragraph," "marketing footer," "chatbot banner," "EU AI Act" — and save each as a Microsoft Word or Google Docs snippet that paralegals can paste in one click. The engagement-letter paragraph goes into the master template once. The chatbot banner goes into the website footer once. The court-filing certification is the only block that gets re-generated per matter, because the model name and verification statement change.
Frequently asked questions
Do lawyers have to disclose use of AI in court filings in 2026?
It depends on the court. Dozens of U.S. district courts and several state courts now have standing orders requiring an AI-use certification. There is no single federal rule yet, but the generator emits the four-element block that satisfies the strictest standing orders enforced so far.
What does ABA Formal Opinion 512 require?
Opinion 512 (July 2024) folds AI use under the existing duties of competence, confidentiality, communication, and reasonable fees. It requires disclosure when AI use is material, when the client reasonably expects to know, or when the engagement letter says so — not on every email.
Are law-firm marketing pages required to disclose AI-generated content?
Under Rule 7.1, yes for the three patterns that trigger it: synthetic attorney headshots, AI-drafted partner posts, and AI-generated testimonials. The marketing preset emits the footer and image label that defuse all three plus a flag for FL, NY, NJ stricter rules.
Does an AI intake chatbot need a disclosure banner?
Yes. Article 50 of the EU AI Act, Rule 7.3 (no live solicitation), and California SB 1001 all apply simultaneously. The chatbot preset emits the opening banner, the first-message disclaimer, and the no-attorney-client-relationship footer in one block.
What goes in a court-filing AI use certification?
Four elements: whether AI was used, the name of the system, that every citation was independently verified by a human attorney, and counsel's signature under the same rules that govern Rule 11. An optional longer narrative covers the role AI played and whether the client was informed.
How does the EU AI Act apply to U.S. law firms?
U.S. firms become subject to the EU AI Act the moment they offer services to EU clients, publish an EU-accessible website using AI features, or process EU client data through an AI system. Article 50 binds on August 2, 2026; penalties scale to €15M or 3% global turnover.