, Nils-Erik Jansson
Contracts with AI vendors: asking beyond standard terms
The standard terms of the large AI vendors cover more than many people think, and less than professional guidance requires. The gap is closed by the addendum the firm or department asks for. This guide sets out the eight points I check against each vendor's terms, using the Swedish Bar Association's model clauses as the yardstick: what the standard terms say, what is missing, and what to do when the vendor says no.
Updated by Nils-Erik Jansson
The method
I put the vendor's commercial terms and data processing agreement next to the list of requirements and the model clauses of the Swedish Bar Association (Sveriges advokatsamfund) [1], and go through them point by point: covered, unclear or missing. Whatever is missing or unclear becomes a draft addendum and a list of questions. The vendor's answers are recorded. What remains open goes into the risk assessment as residual risk, and is handled through the list of matters kept out of the tool. The examples below concern Anthropic's terms as read on 10 November 2026, because that is the vendor the setup covers. The same method applies to any vendor.
| Point | The standard terms say | Ask for |
|---|---|---|
| 1. Confidentiality and ownership | The customer keeps the rights to inputs and owns outputs to the extent the law allows [2]. No reference to legal professional secrecy | An addendum on secrecy and on the information belonging to the firm |
| 2. No training, no anonymisation | No training on customer content under the commercial terms [2]. Exceptions: a right to anonymise the organisation's data for research if the agreement allows it, and handling of flagged content [7] | Deletion of the anonymisation right or confirmation that it is not activated, and a description of how flagged content is handled |
| 3. Authority requests | No such undertaking in the parts I have read [2] [3] | An undertaking in an addendum; otherwise a no or matters kept out, see below |
| 4. Sub-processors and staff | A sub-processor list and a right to object to new sub-processors within a set period, 15 days at Anthropic [3]. Support and security review may take place where group companies and sub-processors are located [4] | The list, monitoring of changes, and a written answer on plain text access for support and when it happens |
| 5. Retention and deletion | Deleted chats are deleted within 30 days, except flagged content and feedback [7]. Custom retention periods only on Enterprise | Confirmation of the deletion routine at the end of the agreement and of what is not deleted |
| 6. Location | Storage in the US and routing by default to several regions; choice of routing location for Enterprise with usage based billing [4] | A written description per component |
| 7. Security reports | Certifications listed in a Trust Center, at Anthropic SOC 2 Type 2, ISO 27001, ISO 42001, CSA Star and NIST 800-171 for Team and Enterprise, with reports on request [5] | The reports, not only the list, and read the exceptions |
| 8. Exit | Export of data in some form, deletion under the retention rules | Written terms on format, timing and what happens to projects, skills and playbooks |
1. Confidentiality and ownership of client information
The model clause: the vendor confirms that the information is covered by the secrecy rules in the Swedish Code of Judicial Procedure (rättegångsbalken), and that all client information, with the rights attached to it, belongs to the firm [1].
The standard terms naturally contain no reference to legal professional secrecy. The point goes into an addendum.
2. No training, no anonymisation
The guidance on generative AI: the tool must not use client information for training or for answers to others [6].
3. Authority requests
The model clause: if a court or authority demands disclosure, the vendor must inform the firm promptly unless mandatory law prevents it. Regardless of that, before anything is disclosed, the vendor must tell the court that the information is covered by legal professional secrecy and oppose disclosure [1]. The guidance expressly refers to the US CLOUD Act for vendors in a US group [1].
Anthropic's data processing agreement and commercial terms contain no such undertaking in the parts I have read [2] [3]. Without that undertaking, the point remains a risk. There are three acceptable outcomes: the vendor gives the undertaking in an addendum; the vendor says no and you accept the risk for certain types of matter; or the matters where the risk is not accepted are kept out of the tool. See the guide on transfer impact assessments.
4. Sub-processors and staff
The model clause: subcontractors and people who see client information are bound by equivalent undertakings, and client information is stored so that it normally cannot be read in plain text, with access logged [1].
5. Retention and deletion
The list of requirements: deletion or return at the end of the agreement, and the firm's immediate and unrestricted access during the term [1].
See the insight on retention in Claude Enterprise.
6. Location
The list of requirements: the vendor must be able to state where client information is, in transit, during processing and at rest [1].
7. Security reports
The list of requirements: ISO 27001 or equivalent, and security work that can be shown through independent audit [1].
8. Exit
The list of requirements: deletion or handover on termination, and exit services [1].
What you have written yourselves, such as playbooks and guidelines, should exist in your own template outside the tool from day one. That is how the setup is built.
When the vendor says no
The guidance on external IT services is clear: if the vendor rejects critical requirements, that goes into the risk assessment, which may lead the firm to choose another vendor [1]. An alternative to switching is to draw a line: the types of matter where the rejected requirement matters are kept out of the tool, and the rest go in. What counts is that the no is in writing and dated, and that the line is set out in the guidelines.
How HOVR Legal works with this
The review of the standard terms against the list of requirements, the draft addendum based on the model clauses and the list of questions for the vendor are part of the data protection package and of the firm package. Negotiation with the vendor beyond the draft and recommendation is yours, and this is stated in what is not included.
Read also legal privilege and AI.
Frequently asked questions
- Do large AI vendors negotiate at all with a small legal department?
- A small legal department has limited bargaining power against a large vendor. What still works is to make the request in writing, get a written answer and record the answer in the risk assessment. A documented no is an acceptable outcome in the assessment. An unanswered request is not.
- Who asks for the addendum, the firm or HOVR?
- The firm or department, as the contracting party. HOVR Legal writes the draft, the commentary on what the standard terms already cover, and the list of questions for the vendor. Negotiation beyond that is work for you or your law firm, and this is stated in what the setup does not include.
- Is the vendor's data processing agreement enough?
- For Article 28 GDPR, usually yes: the large vendors' data processing agreements contain the mandatory terms. What a data processing agreement does not cover is legal professional secrecy, objection to authority requests, and exit on your terms. That is what the addendum is for, and it is why the Swedish Bar Association's guidance on external IT services contains model clauses for exactly this.
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]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, points 4, 9 and 10 and the model clauses in the annex
- [2]Commercial Terms of Service, Anthropic
- [3]Data Processing Addendum, Anthropic
- [4]Where are your servers located?, Privacy Center, Anthropic
- [5]Anthropic Trust Center, certifications and sub-processors
- [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]How long do you store my organization's data?, Privacy Center, Anthropic, read 10 November 2026
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.
- Transfer impact assessment for AI tools, step by step
The six steps of a transfer impact assessment for a US AI tool: map, choose a transfer tool, assess the country, supplement, document, review.
- 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.
- 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.