AI guidance for lawyers in the EU and the UK

A law firm or legal department with work in several countries faces several sets of professional rules on AI at the same time. They say much the same thing, in different words and with different emphasis. This guide sets out what the CCBE, the Solicitors Regulation Authority (SRA), the Law Society, the Bar Council and the Swedish Bar Association require, read in each body's own documents (for the Bar Council, a press release). It covers where they overlap, where they differ, and how one policy with national annexes can cover them all.

Updated by Nils-Erik Jansson

The documents at a glance

Body Document Status Emphasis
CCBE Guides of 2022 and 2 October 2025 Guidance for the European legal profession Five professional duties applied to AI
SRA Risk Outlook of 20 November 2023, warning notice of 17 August 2026 Taken into account in regulation Supervision and competence
Law Society Generative AI: the essentials, updated 30 June 2026 Practice note Checklist; no confidential information in tools the firm does not control
Bar Council Guidance of 30 January 2024 Not formal guidance in the sense of the handbook Verification and judgment
Swedish Bar Association Guidance of 12 June 2025, 15 April 2026 and 10 September 2026 Guidance Consent rule and model clauses

The CCBE: five duties applied to generative AI

The Council of Bars and Law Societies of Europe (CCBE) published a guide on AI tools in 2022 and a guide on generative AI on 2 October 2025. The 2025 guide reads generative AI against five professional duties taken from the CCBE Charter and Model Code [1].

  • Confidentiality. The lawyer is bound by professional secrecy, which covers information about the client and the matter, documents the lawyer prepares and all communication with the client. The lawyer must make reasonable efforts to prevent unauthorised access. Applied to AI, the question is what goes into the tool, under which contract, and who can see it.

  • Professional competence. The lawyer is responsible for the work and must keep up with technical developments that affect practice; the Charter says a lawyer should be aware of the benefits and risks of relevant technology. Applied to AI, that means knowing how the tool works and where it falls short, and reviewing what it produces.

  • Independence. The lawyer must be independent, including from the client and from influences that could distort judgment. Applied to AI, the guide's concern is that the tool's output replaces the lawyer's own assessment.

  • Transparency and information to the client. If it can reasonably be assumed that an informed client would object to, set conditions for or have reservations about the use of generative AI for the purpose, the lawyer should be transparent with the client [1]. This is a test. It does not create a general duty to disclose.

  • Conflict of interest. The guide notes that AI systems trained on, or with access to, confidential information from several clients of the firm can lead to unintended sharing of information or to conflicts [1]. Applied to AI, the question is how project and access boundaries are drawn inside the tool.

The SRA: supervision, competence and a warning notice

The Solicitors Regulation Authority regulates solicitors in England and Wales. Its Risk Outlook report of 20 November 2023 describes hallucination as plausible but incorrect output and mentions cases of invented case law. It makes two points that recur in everything the SRA has said since. First, the high output of an AI system makes effective supervision harder. Second, firms must use AI in ways that protect sensitive information, with particular concern about staff using public tools, data passed to vendors for training, and output that reproduces details of one matter in another [2].

On 17 August 2026 the SRA published a warning notice on the misuse of AI. The SRA takes it into account in its regulatory work. It relies on the Code of Conduct paragraphs on effective supervision of work for clients (paragraph 3.5) and on ensuring that the people a solicitor manages are competent and keep their knowledge of legal, ethical and regulatory obligations up to date (paragraph 3.6). It also relies on the Code for Firms' requirements for supervision systems (paragraphs 4.3 and 4.4), and on Principles 1, 2, 4, 5 and 7: the rule of law and the proper administration of justice, public trust, honesty, integrity and the client's best interests [3].

The Law Society's practice note Generative AI: the essentials, updated on 30 June 2026, turns this into a checklist: follow the SRA Principles and Code, assess which information is suitable to enter, record inputs and outputs, and check the output. It states that misleading a client, the court or anyone else breaches paragraph 1.4 of the Code. Its general advice is not to enter confidential information into generative AI tools whose development and operation the firm does not directly control, because providers may see inputs and outputs and data may be transferred outside the UK [4].

The Bar Council's guidance for barristers, presented on 30 January 2024, makes the same points: verify output because of hallucination and bias, do not treat the tool as a substitute for professional judgment, do not share privileged or confidential information with it, and keep to the procedural rules. The Bar Council points out that the document is not formal guidance in the sense of its handbook [5].

The Swedish Bar Association (Sveriges advokatsamfund) works with three guidance documents: on generative AI (12 June 2025), on external IT services (15 April 2026) and on GDPR (10 September 2026). They are the worked example on this site, because they are the rules that apply to Swedish law firms and the rules the data protection package is built around.

Two features set them apart. The first is a consent rule. No consent is needed if no client information is entered, or if contract and technical measures make the information as secure as in other external IT services. If the advokat (a member of the Swedish Bar Association) judges that risks remain which cannot be addressed by contract or security measures, the client's consent must be obtained, and a clause in the firm's general terms is not enough [6]. The second is model clauses. The guidance on external IT services contains a draft addendum to the vendor contract. It covers professional secrecy, the firm's ownership of client information, storage without plain text access and with logged access, and the vendor's duty to inform the firm and object when a court or authority demands disclosure [7]. More in the guide on legal privilege and AI.

Where they agree

Read side by side, the documents require the same five things. Confidential information is not entered into tools the firm does not control by contract. The lawyer is responsible for the work and must review the output, in particular sources, case law and statutory references. The lawyer must understand the tool well enough to know where it falls short. Supervision must grow with the volume the tool produces. And the client is informed when an informed client would expect it.

Those five points are the core of a policy. They also map directly onto GDPR: processor agreement, transfer assessment, retention and access, and the list of matters kept out of the tool. See the data protection checklist.

Where they differ

Issue Difference
Consent An express requirement in Sweden, generally absent from the UK documents
Regulatory stance The SRA has issued a warning notice and names the paragraphs it applies; the CCBE and the Swedish Bar Association issue guidance
Transparency The CCBE sets a test based on what an informed client would object to; the Law Society and the Bar Council stress the duty not to mislead the court
Contract Sweden provides model clauses; the UK documents state the requirement without a template

None of these differences changes the core of the policy. They change the national annex: what the firm tells clients, what it writes into the vendor contract and what it records for its regulator.

One policy, national annexes

For a firm or legal department in several jurisdictions, the structure that works is a short common policy built on the five points and on GDPR, with one annex per jurisdiction. For Sweden, the annex holds the consent rule and the model clauses. For England and Wales, it holds the SRA paragraphs and the Law Society checklist, with the Bar Council points for barristers. Elsewhere in the EU, it holds the national bar's document, read against the CCBE guide. The data protection documentation is the same for all. Keep each annex to two pages. See the guide on an AI policy for the legal department.

The data protection package is built on GDPR and the EDPB's recommendations, with the Swedish Bar Association's documents as the finished annex. For a firm regulated elsewhere, HOVR Legal writes the common policy and the data protection documentation. The national annex needs a lawyer qualified in that jurisdiction.

Frequently asked questions

Is there a European rulebook on AI for lawyers?
No. The CCBE publishes guides for the European legal profession, but the binding rules are national and sit with the bar or regulator in each country. The CCBE guides are the best statement of the common core, because they are written against the CCBE Charter and Model Code, which most national codes follow [1]. A firm with lawyers in several countries can use the CCBE text as the frame and each national document as an annex.
Does the SRA warning notice create new obligations?
It adds no rules. It states how the SRA applies existing Principles and Code paragraphs to the use of AI, and the SRA takes it into account in its regulatory work. The paragraphs it relies on concern effective supervision of client work and the competence of the people a solicitor manages, together with Principles 1, 2, 4, 5 and 7 [3]. For a firm, that is the difference between advice and expectation.
Where does Sweden differ from the UK?
On consent. The Swedish Bar Association's guidance says that client consent is required where risks remain that cannot be addressed by contract or security measures, and that a clause in the firm's general terms is not enough [6]. The UK documents contain no such rule; they emphasise supervision, competence and the duty not to mislead the court. All of them agree that confidential information does not go into tools the firm does not control by contract.

Statements about Anthropic's terms and products were read on Anthropic's pages on 10 November 2026 and are checked at the start of every engagement.

Sources

  1. [1]CCBE guide on the use of generative AI by lawyers, 2 October 2025, and Guide on the use of AI-based tools by lawyers and law firms in the EU, 2022, 2025 guide section 4; 2022 guide sections 7.3.1, 7.3.3 and 7.3.4
  2. [2]Risk Outlook report: The use of artificial intelligence in the legal market, Solicitors Regulation Authority, 20 November 2023
  3. [3]Warning notice: Misuse of AI, Solicitors Regulation Authority, 17 August 2026
  4. [4]Generative AI: the essentials, The Law Society, updated 30 June 2026
  5. [5]New guidance on generative AI for the Bar, Bar Council, 30 January 2024, press release; the full text is published by Bar Council Ethics
  6. [6]Vägledning om användning av generativ AI i advokatverksamhet, Sveriges advokatsamfund, 12 juni 2025, in Swedish; Swedish Bar Association guidance on generative AI in law practice
  7. [7]Uppdaterad vägledning om användningen av externa IT-tjänster i advokatverksamhet, Sveriges advokatsamfund, 15 april 2026, in Swedish; Swedish Bar Association guidance on external IT services in law practice

Back to AI for legal

More under AI for legal