AI for Lawyers: What the SRA, MinLaw and the Indian Courts Expect in 2026

What solicitors, barristers and advocates in the UK, Singapore and India need to know before their next AI-assisted draft. Regulatory positions, hallucination risk, and a five-step verification workflow.

AI is now in daily use across legal practice, and the regulators have caught up faster than most practitioners realise. This guide covers what solicitors, barristers and advocates in the UK, Singapore and India need to know before their next AI-assisted draft.


Why AI Hallucinations Are a Professional Risk, Not a Technical One

In Ayinde v London Borough of Haringey [2025] EWHC 1383 (Admin) and the companion Al-Haroun v Qatar National Bank, the President of the King’s Bench Division held that a breach of professional standards arising from AI misuse is “extremely serious”, with the full range of sanctions available, including SRA referral and the Hamid jurisdiction.

Singapore has seen a self-represented litigant file submissions containing ChatGPT-fabricated case law. In India, the Supreme Court in July 2026 set aside NCLT and NCLAT orders that had relied on six non-existent AI-generated precedents, holding that a decision tainted by even an “iota” of hallucinated material is no decision in the eyes of the law. Globally, over 1,400 court cases now involve AI-generated hallucinations in filings.


How Large Language Models Actually Work

Most lawyers assume ChatGPT or Claude searches for an answer. It doesn’t. An LLM predicts the next most probable token based on training-data patterns. When it produces a case citation, it isn’t looking anything up. It is generating text that looks like a citation, with a plausible court and year.

That is why citations are the highest-risk legal AI output. Stanford RegLab’s Large Legal Fictions study found general-purpose models hallucinate on specific legal queries 69 to 88 per cent of the time, with the rate rising as questions get more obscure. And the model delivers all of it with total confidence. Fluency is not knowledge.


AI Rules for Solicitors and Barristers in England and Wales

The SRA has issued no AI rulebook and does not plan to. Its position is that the Standards and Regulations 2019 already apply. Firms may use any technology they consider appropriate, “subject to our Principles and Standards”, with the COLP accountable for compliance.

Directly engaged by AI use: Principles 1, 3 and 5 (rule of law, independence, integrity) and Code paragraphs 1.4 (misleading the court, even without dishonest intent), 3.2 and 3.4 (competence), 3.5(b) (supervising juniors), and 6.3 (confidentiality).

For the Bar, the Bar Council’s November 2025 note on generative AI imposes mandatory verification and human oversight, warns against inputting privileged material, and confirms that misleading the court via AI output, even inadvertently, may constitute serious professional misconduct under Core Duties 1, 3 and 5.

Under UK GDPR, uploading client data to an AI platform is processing. Expect Article 28 DPA requirements, transfer safeguards, Article 35 DPIA obligations, and default training-on-inputs settings you must opt out of.


Singapore’s Generative AI Guidance for Lawyers

Singapore has the most developed framework in the common-law world. The MinLaw Guide for Using Generative AI in the Legal Sector (6 March 2026), developed with SAL, the Law Society and SCCA, rests on three principles: professional ethics (accountability cannot be delegated to a tool), confidentiality (avoid free-to-use tools for confidential matters), and transparency (disclose material AI use).

The Law Society of Singapore’s 4R Decision Framework (May 2025) asks four questions before any AI-assisted task: Repetition, Risk, Regulation, Reviewability. If you cannot verify it, you cannot rely on it.

The Courts’ Guide on the Use of Generative AI Tools by Court Users (effective 1 October 2024) permits AI-assisted court documents but requires users to check and verify all output, with full responsibility resting on the user. The PDPA applies to client names, NRIC/FIN numbers and dispute facts entered into overseas-hosted tools.


AI and the Bar Council of India: The Current Position

The BCI has not yet issued formal guidance, but the Supreme Court has directed it to frame rules and disciplinary consequences for filing unverified AI-generated authorities. Meanwhile the courts are acting directly. CJI Surya Kant flagged an “alarming” trend of fictitious citations in February 2026, Justices Datta and Masih rebuked AI-generated fake case law in December 2025, and the Bombay High Court has imposed costs. The DPDP Act 2023 governs client data.


The Productivity Case Is Real

LexisNexis (2025) found lawyers using AI save two to six hours per week. ABA data shows adoption tripling from 11 per cent (2023) to 30 per cent (2024). The rule that separates safe users from sanctioned ones:

AI outputs are always drafts. You are always the professional responsible for the final product.


A Five-Step AI Verification Workflow for Lawyers

  1. Triage. Flag every factual claim, case name, citation or figure you did not supply in the prompt.
  2. Verify every citation. Westlaw UK, Lexis+ or BAILII (UK); LawNet (Singapore); SCC Online or Manupatra (India). If you can’t find it, assume it doesn’t exist.
  3. Verify statutory references. Check legislation.gov.uk, Singapore Statutes Online or India Code. Models misquote and cite superseded versions.
  4. Check currency. Knowledge cutoffs mean recent amendments and judgments are missing.
  5. Sign off personally. It is your work product, not the tool’s.

Get the Full Module

AI Foundations for Lawyers is Module 1 of The AI Bar’s practitioner series, written by practising lawyers, with no technical background assumed. It covers how LLMs work (tokens, context windows, RAG), why hallucinations happen, Claude vs ChatGPT vs Gemini for legal work, the full regulatory position in all three jurisdictions, a data confidentiality checklist with platform settings, the complete verification workflow, 12 worked prompts with a Lawyer’s Check for each, and a one-page Quick-Reference Prompt Card.

Get Module 1: AI Foundations for Lawyers →