Multi-jurisdictional legal AI — Swedish, Norwegian and Danish law without shortcuts
"Are we doing this in Swedish, Norwegian or Danish?" For a lawyer with clients in multiple Nordic countries, that isn't a UI choice but a legal one. Statutes have different names, citation systems are different, the court hierarchies are not the same, and the travaux préparatoires sit in different places.
A global language model with a language switcher can translate the words but not the law. For a legal AI to be usable for Swedish, Norwegian and Danish law, it has to know which jurisdiction the question belongs to and adapt primary sources, citation format, court hierarchy and terminology to match.
This text describes what LexCodex does differently compared to a general chatbot, and why those details matter in client work.
Choice of language and choice of jurisdiction are not the same
Ask a general-purpose LLM in Norwegian about a Swedish contract clause. The answer comes back in Norwegian, but the references point to Norwegian law because the model associates "Norwegian" with "Norwegian law". Ask the other way round: a Swedish-language question about Norwegian employment law gets Swedish LAS reasoning, not arbeidsmiljølov. The same problem appears between Swedish/Danish, Norwegian/Danish and every other pairing.
Handling this correctly requires jurisdiction as a separate input from language, primary sources mapped per jurisdiction, terminology that follows the relevant legal system's vocabulary, and knowledge of where procedural rules differ.
The citation systems are not compatible
This is the most underrated difference. Swedish, Norwegian and Danish lawyers have grown up inside their respective systems and take them for granted. For an AI that needs to produce correct citations, the differences are critical.
| Aspect | Sweden | Norway | Denmark |
|---|---|---|---|
| Statutes | Avtalslagen (1915:218), AvtL § 36 | Avtaleloven (1918), avtl. § 36 | Aftaleloven (LBKG 2016/193), aftl. § 36 |
| Supreme Court case law | NJA 2018 p. 423 | HR-2018-1234-A or Rt. 2018 p. 423 | U 2018.423H or H.D. 12 March 2018 |
| Travaux préparatoires | Prop. 1975/76:81, SOU 1974:83 | Prop. 145 L (2017–2018), NOU 2018:7 | L 145 (2017-18), Folketingstidende |
| Data Protection Authority | IMY (Integritetsskyddsmyndigheten) | Datatilsynet | Datatilsynet |
| Competition Authority | Konkurrensverket (KKV) | Konkurransetilsynet | Konkurrence- og Forbrugerstyrelsen |
| Financial Supervision | Finansinspektionen (FI) | Finanstilsynet | Finanstilsynet |
| Ombudsmen | JO, DO, etc. | Sivilombudet | Folketingets Ombudsmand |
For an AI to produce a correct citation like "Rt. 2018 p. 423" or "U 2018.423H" it needs both the format and access to verified URLs in the relevant primary source. That is the result of per-jurisdiction mapping in the system, not something the model figures out on its own.
How LexCodex handles Swedish, Norwegian and Danish law
Swedish jurisdiction
- Primary sources: legislation, Supreme Court case law, travaux préparatoires and authoritative agency rulings — covered via curated URL mappings.
- Supervisory authorities: IMY, JO, Finansinspektionen, Konkurrensverket.
- Citation format: NJA, RH, FFR, RÅ per Swedish standard.
Norwegian jurisdiction
- Primary sources: lover, forskrifter, Supreme Court case law, proposisjoner, NOU papers and innstillinger — covered via curated URL mappings.
- Supervisory authorities: Datatilsynet, Sivilombudet, Finanstilsynet, Konkurransetilsynet.
- Citation format: HR-NNNN-NNNN-A, Rt. yyyy p. NN, Prop. NN L (yy-yy).
- Bokmål is default. Some legal acts are published only in nynorsk and are handled accordingly.
Danish jurisdiction
- Primary sources: love, bekendtgørelser, Supreme Court case law, lovforslag and betænkninger — covered via curated URL mappings.
- Supervisory authorities: Datatilsynet, Folketingets Ombudsmand, Finanstilsynet, Konkurrence- og Forbrugerstyrelsen.
- Citation format: U yyyy.NNH (Ugeskrift for Retsvæsen), L NNN (folketingsår).
The system prompt is jurisdiction-aware. When the question is asked on the /no/ or /da/ surface, the prompt is constructed so the model knows the question concerns the relevant national law and cites accordingly. That is an explicit framework, not the model guessing.
EU law as common ground
Sweden and Denmark are EU members, Norway is an EEA member via EFTA. GDPR, the EU AI Act, MiFID II, DORA and the AML directives apply equivalently in all three countries. The implementations differ:
- GDPR in Sweden is implemented via dataskyddslagen and sector-specific legislation. Supervision: IMY.
- GDPR in Norway is implemented via personopplysningsloven. Supervision: Datatilsynet.
- GDPR in Denmark is implemented via databeskyttelsesloven. Supervision: Datatilsynet (DK).
- The EU AI Act is directly applicable in Sweden and Denmark as an EU regulation from 2 August 2026.
- The EU AI Act in Norway has been adopted into the EEA agreement in 2024, starts on the same date, and is formally implemented via Norwegian EEA law.
For legal AI used across all three countries, reading the regulation's text is not enough. The system also has to handle how each implementation interacts with national law.
Practical consequences for lawyers
For anyone working with Swedish, Norwegian and Danish clients, or with a cross-border transaction, the differences are concrete:
- Choice of law between a Swedish, Norwegian or Danish party: CISG, Swedish avtalslag, Norwegian avtaleloven or Danish aftaleloven — multiple regimes with different outcomes.
- Confidentiality agreements with employees who relocate: Swedish LAS, Norwegian arbeidsmiljølov and Danish funktionærlov have different rules on non-compete clauses.
- Personal data incidents: reported to IMY, Norwegian Datatilsynet or Danish Datatilsynet depending on where data was processed and where the affected individual is.
- Competition law notification: KKV, Konkurransetilsynet, Konkurrence- og Forbrugerstyrelsen or ESA if EEA aspects are triggered.
Next countries
Sweden, Norway and Denmark all have full LexCodex implementations. We are evaluating Finland and Iceland as the next step. Each country gets a dedicated language version with the same per-jurisdiction architecture.