Sweden
EU-level rules also apply — EU AI Act Art. 4 AI literacy duty applies directly to Swedish law firms deploying AI since 2 Feb 2025. Domestic competence rules in the Code of Conduct do not name AI.
Regulation Tracker · Country · Europe
Sweden
Substantialcivil lawupdated 2026-09-04verified 2026-09-04
EU-level rules also apply — see the European Union record →
Sweden regulates legal AI through bar self-regulation plus directly applicable EU law. The Swedish Bar Association adopted general advice on 13 June 2024 and detailed guidance on 12 June 2025 tying generative AI to the Code of Conduct. Eskilstuna District Court cut a lawyer's fee in January 2026 over four fabricated Supreme Court citations. No court rule requires AI disclosure.
As of 2026-09-04, Sweden has 3 binding rules, official guidance in 9 categories, 0 proposals and 2 positions set only by case law across the 20 categories in the map. Binding rules cover technological competence, horizontal AI statute, mandatory AI training. In 6 categories we found nothing; in 0 the position is not yet determined.
How to read this page
- Binding rule Binding rule means a rule in force from a court, statute, regulator or equivalent authority.
- Official guidance Official guidance means an official non-binding position from a judiciary, regulator, bar or public body.
- Proposed / consultation Proposed / consultation means a public bill, draft rule or consultation that is not yet in force.
- Set by case law Set by case law means no rule was found, but courts have set the position in decisions.
- Nothing found Nothing found means the category was checked and no rule, guidance, proposal or case-law position was found.
- Not yet determined Not yet determined means research did not establish a reliable position yet.
The rules, category by category
The AI Act applies directly. SOU 2025:101 'Anpassningar till AI-förordningen' proposes a new Swedish act with eleven market surveillance authorities, PTS leading, and IMY responsible for Annex III points 6, 7 and 8 — the administration-of-justice category. Sanctions could reach public authorities but not courts in their adjudicative activity.
No AI-specific CLE rule, but bar guidance s.5 says firms should ensure staff receive adequate GenAI training scaled to how the firm uses it, and notes the AI Act may require firms to ensure sufficient staff knowledge. Advokatakademien launched a voluntary AI course in April 2026.
Bar guidance (12 June 2025) s.3: GenAI can cite cases or statutes that do not exist, so a lawyer must critically review output before it enters advice — verifying sources, checking statistics and confirming the applicability of cited law. Advisory liability applies in full regardless of GenAI use (VRGA 5.3, 6.2).
Domstolsverket has internal guidelines for court staff and judges on using public AI tools: classified information must never be entered, and nothing about an undelivered judgment may be shared. Not published as a public practice direction.
Bar guidance s.4: confidentiality [tystnadsplikt] holds regardless of working method; client information may only be entered into GenAI in a secure way, and the tool must not use it for training or to generate output for other users. Firms with information barriers must keep walled data out of shared tools.
Bar guidance s.5 applies the principal's duty under VRGA 7.6.2: a lawyer must take reasonable steps to ensure staff meet good practice when using GenAI, and firms should adopt, publish and regularly update policies on which tools may be used and how output is fact-checked.
Bar guidance s.4: no consent needed if no client information is entered. If client information is used but contractual and technical safeguards match other external IT services, the ordinary assessment applies. Where residual risk cannot be addressed, the client's consent must be obtained — standard terms are not enough.
The board adopted general advice on generative AI models on 13 June 2024; a working group then produced 'Vägledning om användning av generativ AI i advokatverksamhet', dated Stockholm 12 June 2025, with sections on regulatory compliance, advisory liability, confidentiality, and principal liability and office organisation.
Domstolsverket runs AI projects with internal rules: an AI redaction/masking service for confidential data in released documents, and an AI translation tool for judgments. Staff guidelines restrict what may be entered into public tools.
IMY published 'GDPR vid användning av generativ AI' (2024) and a two-part GDPR-and-AI guidance on its innovation portal; Digg and IMY jointly issued guidance on generative AI in public administration. IMY also runs a regulatory sandbox.
Bar guidance sets a vendor checklist: whether the supplier reserves rights over input/output data, whether prompts and client data train the model or reach other users, who is liable for output, and what is disclosed about training data. Lawyers may need IT or cyber-security expertise to assess terms.
Eskilstuna District Court (Jan 2026) held that uncritical AI use producing fabricated authorities weighed strongly against a claimed SEK 2,100 hourly rate, and cut the fee. No bar rule on billing for AI time.
Eskilstuna District Court, judgment reported 13 Jan 2026, criticised counsel for 'AI slop' pleadings citing four non-existent Supreme Court judgments and a misdescribed Contracts Act provision, and reduced the claimed fee. Case number not published in available reports.
No Swedish court rule or practice direction requires parties to disclose or certify AI use in filings as of 4 September 2026; the Code of Judicial Procedure [rättegångsbalken] is silent and Domstolsverket has issued no party-facing direction.
No Swedish rule expressly bars AI from adjudicative reasoning. Domstolsverket's stated position is that AI is used for support tasks only, not decision-making, but this rests on practice and CEPEJ guidance rather than a domestic instrument.
No guidance for self-represented parties using AI found as of 4 September 2026, despite the Eskilstuna case and Swedish legal-press coverage of the issue.
Sweden applies free evaluation of evidence [fri bevisprövning]; no rule or guidance on authenticating AI-generated or deepfake evidence found as of 4 September 2026.
Sweden has no general reservation of legal services (non-advokat legal advisers may practise), so consumer AI legal products raise no unauthorised-practice issue; no ruling found.
The SCC Arbitration Institute had issued no AI guideline or rule as of 4 September 2026; no Swedish ADR body rule found.
Official documents recorded for Sweden (1)
- Vägledning om användning av generativ AI i advokatverksamhet
Swedish Bar Association · revised 2025-06-12 · verified 2026-09-04
Court incidents on the record in Sweden
- Yichang District Court (Hubei)2026-07-21 Warning
Cite this page
SafeLegalAI Legal AI Regulation Tracker, "Sweden" (updated 2026-09-04, verified 2026-09-04), https://safelegalai.com/regulation/country/se (accessed 2026-09-04). Data: CC BY 4.0.
SafeLegalAI is a research publication by Cognesio LLP, not a law firm. Nothing here is legal advice, and no lawyer–client relationship arises from reading it. Rules change; always check the official document linked on each record and take advice on your own situation. Researched and drafted with AI assistance; verified against primary sources and edited by Cognesio LLP. The linked official documents are the record — our summaries are not the law and are not legal advice.
- Advokatsamfundet — Vägledning om användning av generativ AI i advokatverksamhet (PDF)
- Advokatsamfundet — Allmänna råd om generativa AI-modeller (13 June 2024)
- IMY — GDPR vid användning av generativ AI (PDF)
- Sveriges Domstolar — AI masking of confidential data
- Realtid — Eskilstuna tingsrätt AI-fabricated citations (Jan 2026)
- Regeringen — SOU 2025:101 Anpassningar till AI-förordningen