Yes, if the service lets the firm keep client confidentiality. The Swedish Bar Association (Advokatsamfundet) sees no real obstacle to law firms using cloud services, but requires that the duty of confidentiality and the protection of client information are upheld. On the free version of ChatGPT, content is used to train the models by default, and then the answer is in practice no.
What the law says
Under chapter 8, section 4 of the Swedish Code of Judicial Procedure, an advokat must keep confidential what they learn in their practice when good professional conduct requires it. A breach can be tried under chapter 20, section 3 of the Criminal Code, but only the Chancellor of Justice may prosecute, and only when the public interest requires it. In practice a breach is normally handled as a disciplinary matter by the Bar Association.
The law does not mention AI. So the question is the same as for any other service: can the firm show that the client's information is not disclosed?
What the Bar Association says
- June 2024: the board adopted general advice on generative AI models in legal practice.
- 2025: the advice was replaced by guidance on the use of generative AI. Among other things it says tools should only be used if they ensure client information is not disclosed in breach of confidentiality, that the lawyer should review the tool's terms, and that firms should have internal guidelines.
- 3 February 2026: the association presented framework agreements with four legal AI suppliers: Blendow Lexnova, Legora, Juno and JP Infonet.
- 28 April 2026: the guidance on external IT services was updated. The conclusion is the same as in 2011 and 2019: no real obstacle to cloud services, as long as confidentiality and the other professional duties are kept. The new version points to the geopolitical situation and adds a supplementary agreement with suggested clauses for supplier contracts.
What it means for ChatGPT
| Plan | Trained on your content? | Price |
|---|---|---|
| Free, Go, Plus, Pro | Yes, by default. It can be switched off. | From SEK 0 |
| Business | No, by default | SEK 180 per user a month, billed yearly |
| Enterprise | No, by default. Data can be stored in one of ten regions, the EU among them. | On quote |
A private ChatGPT subscription on a lawyer's own card is the wrong tool for client material even with training switched off: the firm has no contract and no control. Business and Enterprise are firm contracts with training off from the start. But OpenAI is a US company, and that decides which law can reach the data, wherever it is stored. The note on the CLOUD Act and FISA 702 covers that.
A checklist for the firm
- Which plan, and whose contract? A firm contract, not a private subscription.
- Is anything trained on what you enter? No should be the default, not a choice each user must make.
- Where is the material stored, and which country's law applies to the provider and its owners?
- Is it logged who opened what? Without a log the firm cannot show who saw the client's information.
- Are there internal guidelines, as the Bar Association recommends? Which matters may go in, and what must never be pasted?
- Does the supplier contract carry clauses in line with the supplementary agreement in the 2026 guidance?
Quince AI builds Egen, an AI workspace on Swedish servers with no US company in the chain. The checklist holds whichever tool you choose.
Sources
- Rättegångsbalken 8 kap. 4 §, via Advokatsamfundet
- Justitiekanslern, beslut 2021-02 om advokaters tystnadsplikt
- Advokatsamfundet, allmänna råd om generativa AI-modeller, 18 juni 2024
- Advokatsamfundet, vägledning om användning av generativ AI i advokatverksamhet (2025)
- Norstedts Juridik om Advokatsamfundets ramavtal om AI-tjänster, februari 2026
- Advokatsamfundet, uppdaterad vägledning om externa IT-tjänster, 28 april 2026
- ChatGPT, priser och planer, kontrollerat 24 september 2026