, Nils-Erik Jansson
Transfer impact assessment for AI tools, step by step
The GDPR guidance of the Swedish Bar Association (Sveriges advokatsamfund) sets out a transfer impact assessment in six steps for law firms. The same steps are useful when a legal department puts personal data into Claude, ChatGPT or another tool from a US vendor. That applies even if the data is said to be stored in the EU, because access from a third country is a third country transfer. Here are the steps, what they mean for an AI tool in particular, and what the conclusion looks like.
Updated by Nils-Erik Jansson
Why AI tools in particular
A transfer impact assessment for an AI tool is the same as for any cloud service. The difference is that the content is richer. A case management system holds names and file numbers. A prompt holds the whole contract, the whole dispute and the lawyer's own assessment. That makes steps three and four more important than usual, and it makes the list of matters kept out of the tool one of the results.
The Swedish Bar Association's GDPR guidance says the assessment is made in a six-step procedure, with a duty for the firm to carry out a transfer impact assessment [1]. The steps come from the European Data Protection Board's Recommendations 01/2020 [2]. This is how they look for an AI tool.
Step 1: map the transfers
Which personal data goes into the tool, about whom, and where does it go? For an AI tool the answer is more than one country. Anthropic states that storage is in the US, that traffic may by default be routed to the US, Europe, Asia and Australia, and that support and security review can take place where group companies and sub-processors are located [4]. The sub-processor list is part of the mapping. The result of step one is a table of data, data subjects, recipients, country and purpose. For an AI tool it starts like this, with what Anthropic states.
| Transfer | Recipient | Country | Data, data subjects and purpose |
|---|---|---|---|
| Storage | Anthropic | US | You fill in |
| Default traffic | Anthropic | US, Europe, Asia and Australia | You fill in |
| Support and security review | Group companies and sub-processors | Where they are located | You fill in |
| Sub-processors | As in the sub-processor list | As in the list | You fill in |
| Integrations with contract archive, email and document management | Each system | You fill in | You fill in |
Integrations with other systems (contract archive, email, document management) give the tool access to more than what is typed into the prompt, and belong in the table.
Step 2: choose a transfer tool
An adequacy decision, standard contractual clauses, binding corporate rules or one of the derogations in Article 49. For a vendor that is not certified under the Data Privacy Framework, that means the standard contractual clauses in the data processing agreement. Anthropic's DPA is based on the 2021 clauses, modules 2 and 3 [5]. Check which modules are completed, who is exporter and who is importer in the step from the Irish company to the US, and that the annexes actually describe your processing. The Article 49 derogations cannot carry an ongoing vendor relationship.
Step 3: assess whether the transfer tool holds in the destination country
This is the core. The question is whether law and practice in the destination country can prevent the importer from complying with the clauses, especially when public authorities seek access. For the US that means FISA 702 and the Cloud Act, which the Swedish Bar Association's guidance on external IT services points to: a US court can order a US company to disclose information even if it is stored in the EU [6]. IMY (the Swedish Authority for Privacy Protection) writes that the measures the US has taken since Schrems II apply to all transfers to US companies, whatever the transfer tool, and can be taken into account in the assessment [3].
Document what the vendor says it does on a request, what it does not promise, and how likely a request is for your data in particular. Where the likelihood is low but the consequence high for some of the matters, the list of matters kept out of the tool deals with them.
Step 4: supplementary measures
If step three shows a risk, it must be met with technical, contractual or organisational measures. Under the recommendations, technical measures are the only ones that can prevent access by public authorities. For an AI tool that has to read plain text to work they are limited: encryption in transit and at rest exists, encryption during processing does not. Contractual measures are a supplementary agreement requiring information and objection when an authority makes a request, such as the Bar Association's model clauses [6]. Organisational measures are de-identification before data goes in, short retention periods, and the list of matters that do not go in at all.
Be honest in the documentation about what the measures cannot do. An assessment that says everything is solved by a contract clause will not survive scrutiny.
Step 5: formal steps
The standard contractual clauses need no approval. They must, however, be signed in the right version, with the annexes completed and any supplementary agreements in place before the first data goes in. An impact assessment under Article 35 is carried out where the processing requires it. The data protection officer should have read the assessment.
Step 6: review
Once a year, and when something changes: the vendor changes sub-processors (Anthropic's DPA gives 15 days to object [5]), the legal position changes, or the department starts using the tool for new types of matter. The GDPR guidance also requires the firm to have a plan for alternatives or a change of vendor if the adequacy decision for the US falls [1]. That plan is one page: what happens to the data, how quickly, and which tool takes over.
Where the assessment ends
The assessment ends in a yes for the types of matter where the risk can be met, with a supplementary agreement or a documented request for one, with short retention periods, and with a list of matters that do not go in. That list is the most concrete result, and it belongs in the AI policy.
How HOVR Legal works with this
The transfer assessment is one of five parts of the data protection package. HOVR Legal delivers it in your template and goes through it with your data protection officer, and it carries over to the next tool. For Claude it is part of the setup, with Anthropic's material already reviewed.
Frequently asked questions
- Is a transfer impact assessment needed if the vendor stores data in the EU?
- Yes, if the data can be accessed from a third country. The GDPR guidance treats access from outside the EU/EEA as a transfer, for example when support staff get remote access, so the storage location does not settle the question [1]. Anthropic states that traffic can be routed to several regions and that support and security review can take place where group companies and sub-processors are located [4]. There is then a transfer to assess.
- Is it enough that the vendor is certified under the Data Privacy Framework?
- If the vendor is certified, the transfer can rest on the adequacy decision for the US. The assessment becomes simpler but is still needed: the guidance says the firm should have a plan for alternatives if the adequacy decision falls [1]. When I searched the register, Anthropic was not among the active participants, see the insight on the DPF register. Then the standard contractual clauses apply, and the assessment is mandatory.
- Who makes the assessment, us or the vendor?
- You do. The vendor provides material: where data is processed, which sub-processors there are, what steps it takes on an authority request. Responsibility for the conclusion lies with the controller, that is the department or the firm. That is also why the assessment should be in your own template, and not only exist as a document from the vendor.
Sources
- [1]GDPR-vägledningen, Sveriges advokatsamfund (Swedish Bar Association), 10 September 2026 (in Swedish), sections 12.2 and 12.3
- [2]Recommendations 01/2020 on measures that supplement transfer tools, version 2.0, European Data Protection Board, 18 June 2021
- [3]Överföringar av personuppgifter till USA, IMY (in Swedish)
- [4]Where are your servers located?, Privacy Center, Anthropic
- [5]Data Processing Addendum, Anthropic
- [6]Uppdaterad vägledning om användningen av externa IT-tjänster i advokatverksamhet, Sveriges advokatsamfund (Swedish Bar Association), 15 April 2026 (in Swedish)
Insights on the topic
Back to AI for legal
More under AI for legal
- Data protection package for AI tools in legal work
Fixed-scope engagement: transfer impact assessment, impact assessment where needed, draft addendum, policy and training. For the AI tool you use or consider.
- AI operating model for the legal department
Fixed-scope engagement for legal departments. You get an operating model for how AI is used, governed and followed up, whatever the tool.
- AI literacy training for lawyers: two hours
Two hours of training for your legal department or firm on your own tools and documents, documented as a measure under Article 4 of the AI Act.
- Data protection checklist for generative AI in legal work
A checklist for law firms and legal departments: five questions to answer before the first client data goes into an AI tool, and what counts as a pass.
- AI policy for the legal department: what it should contain
Outline of an AI policy for a legal department or law firm: permitted tools, what may be entered, review, responsibility, excluded matters and training.
- AI policy template for the legal department
An editable AI policy template for a legal department or law firm in nine sections, with comments on what to adapt and why.
- The EU AI Act for in-house counsel: what applies to users
What the EU AI Act requires of a legal department that uses generative AI: AI literacy since 2025, transparency since 2026 and when the high-risk rules apply.
- DPIA for generative AI: when it is required and how
When a legal department or law firm must carry out a DPIA under Article 35 GDPR for an AI tool, using the Swedish regulator IMY's nine criteria.
- Personal data in prompts: legal basis and information
When a lawyer writes about an employee, a counterparty or a customer in an AI tool, personal data is processed.
- AI incident response for legal teams
What a legal department does within 72 hours when a member of staff has entered data into an AI tool that should not have gone in.
- Anonymise before prompting: what works
How a lawyer removes personal data before text goes into an AI tool, and the difference between anonymisation and pseudonymisation.
- Swedish MBL and AI tools: must you negotiate first?
Swedish law: when an AI tool triggers the duty to negotiate under section 11 MBL, when informing under section 19 is enough, and with no collective agreement.
- Shadow AI in the legal department: one week of work
Staff use ChatGPT and Claude on private accounts for work. The legal department's week: map it, stop the worst, offer an approved tool, write the rule.
- Legal privilege and AI: the Swedish Bar guidance
What the Swedish Bar Association's guidance on generative AI requires for confidentiality: tools that do not train on client data, and terms read in full.
- Contracts with AI vendors: asking beyond standard terms
What a legal department or law firm asks of an AI vendor beyond the standard terms: authority requests, sub-processors, deletion, anonymisation, exit.
- On-premise or cloud AI for legal work
When a legal department or law firm should run AI locally and when the cloud is right: what local operation solves, what it costs, what it leaves open.
- Reviewing an AI answer in ten minutes: five steps
How a lawyer reviews an answer from Claude, Legora or ChatGPT before using it: sources, case law and statutes, figures, what is missing, and tone.
- AI literacy training for the legal department
What a two-hour training for lawyers using generative AI covers: what the tool does and does not do, your rules, practice on your own documents.
- AI guidance for lawyers in the EU and the UK
What the CCBE, the SRA, the Law Society, the Bar Council and the Swedish Bar Association require of lawyers using generative AI, and where they differ.
- The general counsel's AI report to the board: two pages
What the board needs to know about the company's AI use once a year, and what the general counsel can answer for.