By The AI Bar  · 

AI Ethics for Lawyers: What the Rules Require and What the Courts Are Punishing

No regulator bans AI in legal practice — but every one holds you personally responsible for what you file. Here's what the rules say in India, England and Wales, and Singapore, and what the enforcement record shows.

Answer up front: no regulator in India, England and Wales or Singapore bans AI in legal practice. Every one of them holds you personally responsible for what you file. The sanctions are already real: by April 2026, over 600 documented AI-hallucination cases had been filed in courts globally, implicating 128 lawyers, with new cases reported at two to three per day.

The Four Ethical Pillars

Competence. Understand the tool. Know its data cutoff, its jurisdiction coverage and its hallucination rate. Ignorance is not a defence.

Confidentiality. Anonymise before every prompt. Check data-retention terms. Confidentiality is unqualified, not a reasonable-steps obligation.

Supervision. Read everything before it leaves your desk. A partner’s signature is a partner’s responsibility, whatever generated the text.

Honesty to the court. Verify every citation in the original database before filing. If you cannot find the case, do not cite it.

What the Rules Say in Each Jurisdiction

India: the Advocates Act 1961 and BCI Rules impose duties of honesty to the court and to the client. There is no AI-specific rule yet, which means the general duties apply in full.

England and Wales: the SRA Code and Bar Council guidance apply unchanged. SRA compliance guidance states it should always be made clear to clients where they are interfacing with AI.

Singapore: the clearest framework of the three. MinLaw’s GenAI Legal Sector Guide, anchored in rule 5 of the Professional Conduct Rules 2015, advises practitioners to consider informing clients of GenAI use where it is substantial, affects cost, or involves data handling the client may not accept.

The regulatory gap in India and the UK is not a permission slip. It means enforcement runs through existing conduct rules, which are broader, not narrower.

The Enforcement Record

  • R (Ayinde) v London Borough of Haringey [2025] EWHC 1383 confirmed the English courts treat AI-hallucinated authorities as a serious matter.
  • Johnson v. Dunn (N.D. Alabama, July 2025): three Butler Snow attorneys disqualified and referred to state bars in every jurisdiction where they were licensed.
  • Supreme Court of India, 17 February 2026: the Chief Justice flagged an “alarming” trend of fabricated citations, including a case titled “Mercy v. Mankind”.
  • Nathwani (Supreme Court of Victoria, August 2025): the first Australian practising certificate varied for AI misuse.

Five principles emerge from the record: not knowing the AI was wrong is not a defence; the duty to verify is personal and non-delegable; sanction severity is escalating from costs orders to disqualification and licence variation; enforcement is now global, not a US phenomenon; and prompt withdrawal plus candid disclosure mitigates but does not eliminate sanction.

The Moffatt Principle: There Is No “The AI Told Me So”

Moffatt v. Air Canada 2024 BCCRT 149 rejected the argument that an AI chatbot gave independent advice for which the operator was not responsible. AI tools are instruments, not agents. They generate content you adopt as your own professional output, and every consequence — malpractice liability, disciplinary sanction, costs order — attaches to the lawyer who signed it.

The AI Stop/Go Framework

Go, with review:

  • First-draft generation where you read every word
  • Document summarisation verified against the original
  • Billing narratives accuracy-checked before posting
  • Research orientation with every citation verified in the primary source
  • Compliance first drafts with regulatory references verified

Stop, human only:

  • SAR/STR filing decisions
  • Court deadline calculation without manual verification
  • Direct communication with clients or courts without review
  • Processing confidential client data in unsecured tools
  • Substituting AI judgment for professional judgment in advice

Governance: The Seven-Item Minimum

  1. Written AI use policy, signed by all fee earners.
  2. Tool inventory with every approved tool documented.
  3. Task-specific written verification protocol, filed per matter.
  4. Engagement letter updated to disclose AI use.
  5. Professional indemnity policy reviewed for AI coverage.
  6. All fee earners trained on AI conduct obligations.
  7. Annual governance review scheduled.

If your firm has none of these, start with the verification protocol. It is the one that prevents the sanction.

Frequently Asked Questions

Can lawyers use AI ethically? Yes. No regulator in India, the UK or Singapore prohibits it. What each requires is competence in the tool, protection of client confidentiality, supervision of output, and personal verification of anything filed with a court.

What happens if a lawyer files AI-hallucinated citations? Sanctions to date include costs orders, wasted-costs findings, referral to bar associations, disqualification from the matter, and variation of a practising certificate. Courts have consistently rejected inadvertence as a complete defence.

Do I have to tell clients I used AI? No jurisdiction mandates universal disclosure yet, but the direction is clear. Singapore’s GenAI Legal Sector Guide expects it where use is substantial or affects cost or data handling; the SRA requires clarity where clients interface with AI directly. Best practice is a short standing statement in your engagement letter.

Who is responsible for AI output in a law firm — the associate or the partner? Both, on different footings. The associate owes a personal duty to verify; the supervising partner owes a supervision duty that is not discharged by delegation. The signature carries the liability.

Does professional indemnity insurance cover AI errors? Do not assume so. Insurers are actively repricing and adding AI-specific questions and conditions. Review your policy wording and disclose your AI use and governance controls at renewal.

This article is a summary. The full treatment lives in Module 8: Ethics, Risk & the Future — the final module of The AI Bar series, covering the four pillars in depth, the professional conduct frameworks for India, England and Wales and Singapore, the complete enforcement casebook, malpractice and PI exposure, supervision and disclosure obligations, agentic AI, the future of legal careers, a full AI governance framework, and 12 worked prompt examples you can use immediately.