AI Agents for Lawyers

Questions

Is It Ethical for Lawyers to Use AI? The Rules in the US, UK and EU

Yes, every major bar permits lawyers to use AI. It becomes unethical when output is filed unverified or client data is exposed. US, UK and EU rules in 2026.

By Dr. Niklas SchmidtUpdated 13 November 20266 min read
On this page
  1. The six duties every US opinion applies
  2. What the UK and European bars add
  3. Failure one: output nobody verified
  4. Failure two: client data in a tool that keeps it
  5. The grey areas: disclosure, billing, recording
  6. An ethics checklist for this week

Yes. As of September 2026 no bar in the United States, England and Wales, Germany or Austria treats using AI as misconduct, and the National Association of Criminal Defense Lawyers argues that not using it can be. The ethics turn on how, not whether. Two failures account for nearly every disciplinary case: filing output nobody verified, and putting client information into a tool that keeps it.

The six duties every US opinion applies

The ABA's Formal Opinion 512 (29 July 2024) created no new rule. It applied six existing ones, and the 17 state and local opinions issued since follow the same map. Only Alabama (Formal Opinion 2026-01) has added one in 2026; California went further, approving AI-specific comments to six Rules of Professional Conduct on 13 March 2026 and Practical Guidance covering generative and agentic AI on 14 May 2026 (Esquire Solutions).

DutyModel RuleWhat it requires when you use AI
Competence1.1Understand the tool's benefits and risks; Texas Opinion 705 calls it "a reasonable and current understanding of the technology"
Confidentiality1.6Informed consent where a self-learning tool may expose client data; California's proposed comment treats exposure as "revealing" where there is "a material risk of unauthorized access, retention, or use"
Communication1.4Tell the client where AI affects the representation, the risk or the fee
Candour to the tribunal3.3Every citation read and verified by the lawyer responsible for the filing
Supervision5.1 and 5.3Policies, training and review for lawyers and non-lawyer staff
Reasonable fees1.5Bill actual time, never time saved; subscriptions are generally overhead

The Opinion 512 explainer and the state-by-state guide carry the detail.

What the UK and European bars add

The SRA's Warning Notice on misuse of AI (17 August 2026) records 42 misuse reports between July 2025 and July 2026 and sets the English test: a solicitor must understand a tool "well enough to review its output critically". It builds on Ayinde v Haringey [2025] EWHC 1383 (Admin), where Dame Victoria Sharp P listed sanctions running from public admonition to contempt and police referral, and addressed heads of chambers and managing partners directly (Burges Salmon). The SRA notice and Ayinde pages go through both.

On the continent the emphasis shifts to the client. The CCBE's guide on generative AI (2 October 2025) says inform clients and let them object, and flags hallucination and data-retention risks (eucrim); its Technical Guide of 27 March 2026 extends to agentic AI. The German BRAK's Hinweise (December 2024) state that tools which store or forward client data for model training breach § 43a Abs. 2 BRAO (BRAK). The Austrian ÖRAK's four practice rules fit on a card: client data stays on EU servers, disclose AI use, the lawyer keeps final responsibility, no client data for training (Anwaltsblatt). Over all of it sits Article 4 of the EU AI Act: since 2 February 2025 any organisation whose staff use tools like ChatGPT must take measures to support AI literacy, national authorities have supervised it since 2 August 2026, and the Commission's FAQ asks for an internal training record rather than certificates. The Article 4 page and the ÖRAK, BRAK and CCBE page have the specifics.

Failure one: output nobody verified

Damien Charlotin's database held 2,041 court decisions involving AI hallucinations on 14 September 2026: 811 involving lawyers, 32 involving judges (AI Hallucination Cases). Judge Rankin's line in Wadsworth v. Walmart (D. Wyo., February 2025) is the whole rule: "a fake opinion is not existing law". The sanctions have escalated from Mata v. Avianca's $5,000 in 2023. Couvrette v. Wisnovsky (D. Or., 12 December 2025) cost $110,204.38, largely because counsel quietly refiled amended briefs instead of confessing (ABA Journal). The Nebraska Supreme Court imposed an interim suspension (order 15 April 2026) on a lawyer whose divorce-appeal brief had 57 of 63 citations defective. In Johnson v. Dunn (N.D. Ala., 23 July 2025) Butler Snow partners were removed from the case and reprimanded. Sullivan & Cromwell apologised to a bankruptcy judge in April 2026. Beus Gilbert PLLC v. BYU (D. Utah, 9 September 2026) added two AI-ethics CLE courses to a $3,000 fine.

The failure is not ignorance. Scientific American (May 2026) attributes the plateau of 350–400 new decisions a quarter to "cognitive surrender", and reports a Nature study finding that people who like technology perform worse when told advice comes from AI (Scientific American). The cure is procedural, not attitudinal: the citation-checking protocol and the sanctions case list.

Failure two: client data in a tool that keeps it

In United States v. Heppner (S.D.N.Y., opinion 17 February 2026) Judge Rakoff held that a defendant's research memos generated in consumer Claude were never privileged: "Claude is not an attorney", and the consumer terms permit logging and disclosure (Harvard Law Review). The UK Upper Tribunal said in UK v SSHD [2026] UKUT 81 (IAC) that pasting client letters into public AI tools puts them "in the public domain". Judge Stein's affirmation (5 January 2026) of an order producing 20 million de-identified ChatGPT logs confirmed that chats are discoverable ESI.

The mechanism is the plan. Consumer tiers of ChatGPT and Gemini train on inputs unless you opt out; Google's own help page tells consumer Gemini users not to enter confidential information a human reviewer might see; Claude Free, Pro and Max users had to choose. Business, Enterprise and API tiers do not train by default and offer zero data retention by contract. Under BRAK's reading of § 43a BRAO and California's proposed Rule 1.6 comment, the consumer tier is not a preference but a breach. The privilege page covers Heppner, Warner v. Gilbarco and the logs litigation.

The grey areas: disclosure, billing, recording

Three questions have no single answer yet. Disclosure to courts: Illinois requires none in pleadings, while over 30 of 94 US federal district courts had disclosure or certification requirements by April 2026 and Judges Newman and Boyko in Ohio prohibit generative AI except legal search engines and Microsoft products. Billing: Texas Opinion 705 forbids charging for time "saved" (Texas Center for Legal Ethics), Virginia LEO 1901 allows an unchanged value-based fee, California forbids marking up AI costs without informed written consent; the billing page reconciles them. Recording: NYC Bar Formal Opinions 2025-6 and 2026-2 require the consent of all parties to AI transcription, forbid surreptitious recording and set the default to off (NYC Bar).

Insurance is the quiet fourth. No major US professional liability carrier had filed an explicit AI exclusion as of 5 May 2026, but application forms now ask whether the firm lets AI draft documents, and Armilla at Lloyd's sells affirmative cover that names lawyer hallucinations.

An ethics checklist for this week

  • Plan tier confirmed: no training on inputs, data processing agreement signed, zero data retention for privileged material.
  • Dated screenshot of the settings on file. This is the SRA's "understand the tool" evidence and the AI Act Article 4 training record in one.
  • Verification rule written down: every citation opened in the primary source by the lawyer who signs; log kept on the matter.
  • One sentence on AI use in the engagement letter, with the client's right to object.
  • Fees: actual time or an agreed fixed fee; no mark-up on subscriptions.
  • Recording: consent obtained first; transcription off by default.
  • Supervision: a written policy and one recorded training hour per person, including staff.

Module 6 of the AI Lab for Lawyers course works through each item against the actual sanction orders and settings screens. The bar has answered the ethics question; what remains is whether the firm has written the answer down.

Frequently asked questions

Is it unethical for a lawyer to use ChatGPT?

No. Using ChatGPT or any generative AI is permitted under ABA Formal Opinion 512, the SRA's 2026 warning notice, the CCBE guide and the Austrian and German bar guidance. It becomes unethical when the lawyer files what the tool invents without checking it, or puts identifiable client information into a consumer plan that trains on inputs or keeps logs that courts can order produced.

Can a lawyer be disciplined for using AI?

Yes, for misusing it. As of September 2026 courts have issued fines up to $110,204, removed lawyers from cases, referred them to bars and, in Nebraska, imposed an interim suspension after 57 of 63 citations in a brief were defective. In England the Divisional Court in Ayinde referred lawyers to the SRA and BSB and listed sanctions up to contempt and police referral.

Do lawyers have a duty to use AI?

Not yet as a general rule, but the argument is being made. The National Association of Criminal Defense Lawyers said in August 2026 that defenders have an ethical duty to use AI because prosecutors already do, and in-house commentators predict that not using it will be treated as malpractice. Forty US states have adopted the technology-competence comment, which already requires understanding the tools.

What are the ABA rules on lawyers using AI?

ABA Formal Opinion 512 of 29 July 2024 applies six existing Model Rules: competence, confidentiality, communication, candour to the tribunal, supervision and reasonable fees. It requires informed consent where a self-learning tool may expose client data, allows billing for actual time but not time saved, and treats general AI subscriptions as overhead. Seventeen state and local jurisdictions have issued their own opinions on the same lines.

Put this into practice

AI Lab for Lawyers: a live, hands-on course for practising lawyers

4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.

Reserve a seat · €200 Maven money-back guarantee · certificate of completion

Regulation & Ethics

ABA Formal Opinion 512 Explained: What US Lawyers Must Do When Using AI

ABA Formal Opinion 512 applies six Model Rules to generative AI: competence, confidentiality, communication, candour, supervision and fees. What each duty requires in 2026.

8 min read

Regulation & Ethics

The SRA's 2026 Warning Notice on AI Misuse: What UK Solicitors Must Know

The SRA AI warning notice of 17 August 2026 followed 42 misuse reports in a year. Solicitors must understand a tool well enough to review its output critically.

8 min read

Regulation & Ethics

AI Rules for Austrian and German Lawyers: ÖRAK, BRAK and CCBE Guidance

ÖRAK's four AI rules and § 40 Abs 3 RL-BA bind Austrian lawyers; BRAK's § 43a BRAO line binds German ones; CCBE guides cover both. With a bilingual checklist.

11 min read

Regulation & Ethics

Do You Have to Tell Clients You Use AI? Disclosure Rules and Practice

When lawyers must disclose AI use to clients under ABA 512, California, CCBE and ÖRAK rules, when consent is required, and a sample engagement-letter clause.

10 min read

Questions

Can Lawyers Use ChatGPT for Legal Work? Yes — With These Conditions

Lawyers can use ChatGPT for legal work on a non-training plan if they verify every citation, tell clients and bill honestly. The rules as of September 2026.

6 min read

Regulation & Ethics

How to Write a Law Firm AI Policy (With a Template Outline)

A law firm AI policy lists approved tools by tier, classifies data, makes verification personal, covers disclosure, billing and training, and limits agents.

9 min read