AI due diligence
AI due diligence: what to ask a vendor
Before a firm lets an AI tool near client information, it should check how the tool handles that information. These are the questions the framework asks, with what law societies and regulators have published on the same topics.
Quotes last checked on their source pages on 3 November 2026.
How to read this page
Each group below opens to show its questions. A question is the framework’s own wording. Under it are short quotes from the bodies that publish guidance, each with the issuer, the title, the date and a link to the page it came from.
A quote shows what that body wrote. It is not our summary of its view, and it does not say what answer passes. Some questions have no quote because none of the sources we checked speaks to them directly.
Permitted use and the firm’s duties
Where the requirements come from, and whether client information may be used at all.
The 2 questions in this groupShow the questions
Where the requirements come from
Record the actual sources driving the assessment; requirements can come from several places.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“follow professional obligations under the SRA Code of Conduct, SRA Standards and Regulations and SRA Principles”
- Law Society of England and Wales: Generative AI – the essentials
“follow any relevant guidance, including from regulators and the court”
Authority and permitted use
Confirm disclosure and use are authorised for the stated tasks. Consent, privilege and professional obligations are separate matters to assess.
What the sources say
- Law Society of NSW, Legal Practice Board of WA and Victorian Legal Services Board + Commissioner (joint statement): Statement on the use of artificial intelligence in Australian legal practice
“If lawyers use commercial AI tools with any client information, they need to carefully review contractual terms to ensure the information will be kept secure.”
- Solicitors Regulation Authority: Compliance tips for solicitors regarding the use of AI and technology
“keeping the affairs of the current and former clients confidential”
- American Bar Association: Formal Opinion 512: Generative Artificial Intelligence Tools
“merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient”
Where the data goes
Storage, model, processing, backup, log and support locations, and cross-border transfers.
The 8 questions in this groupShow the questions
Primary storage location
Where files, prompts, outputs and indexes are stored at rest.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“how and where the data is stored”
- New Zealand Law Society: Generative AI guidance for lawyers
“the geographic location of data storage”
Model inference location
Where model computation takes place. Check routing settings and the exact model.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“where the data is being processed”
Other processing location
Include OCR, extraction, indexing, CPU processing and coordination outside model inference.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“where the data is being processed”
- Law Society of England and Wales: Generative AI – the essentials
“who is processing the data”
Fallback and disaster recovery location
Check overflow, failover and outage behaviour as well as the normal route.
None of the sources checked speaks to this question directly.
Logs and safety-monitoring location
Include logs containing prompts, outputs or derived sensitive information.
None of the sources checked speaks to this question directly.
Backups and replicas location
Confirm replicas, backups and recovery copies, not only the primary database.
None of the sources checked speaks to this question directly.
Support and staff-access location
Location of remote access matters even where the file stays in its primary region.
None of the sources checked speaks to this question directly.
Cross-border processing basis
Evaluate the actual transfer obligations for the jurisdictions and information involved.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“As many generative AI companies are located outside of the UK, data may be transferred outside UK borders”
- New Zealand Law Society: Generative AI guidance for lawyers
“The data inputted may also be transferred out of New Zealand to AI companies located overseas.”
Who handles it
The model provider, subprocessors, supplier staff, connectors and web search.
The 5 questions in this groupShow the questions
Model-provider arrangement
Identify who processes content for inference and under which contract.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“who is processing the data”
- Law Society of England and Wales: Generative AI – the essentials
“carry out due diligence (including supplier due diligence) on the AI tools you use”
- Solicitors Regulation Authority: Compliance tips for solicitors regarding the use of AI and technology
“You should undertake due diligence to ensure that these platforms have been designed so that you are not in breach of your obligations.”
Subprocessor register
Include model hosting, OCR, search, support, analytics, identity and any other recipient of relevant information.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“Careful consideration needs to be taken when client data is to be shared with a third-party AI vendor.”
- Law Society of England and Wales: Generative AI – the essentials
“who is processing the data”
Supplier human access
Separate customer-authorised support, safety review, legal access and routine operations.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“You should be cautious that generative AI companies may be able to see your input and output data.”
- New Zealand Law Society: Generative AI guidance for lawyers
“Be aware that the AI provider may well be able to see your input data and the outputs.”
- Law Society of England and Wales: Generative AI – the essentials
“who has access to the data”
Connectors and external services
Every external service may have separate data handling, permissions and retention.
None of the sources checked speaks to this question directly.
Public web search
Derived search terms can reveal confidential facts even when the full document is not sent.
None of the sources checked speaks to this question directly.
How the data is used and kept
Training, feedback, retention and deletion.
The 4 questions in this groupShow the questions
Training and service improvement
Distinguish foundation-model training from customer fine-tuning, safety review and other service improvement.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“establish whether the generative AI tool is a closed system within your firm’s boundaries or also operates as a training model for third parties”
- New Zealand Law Society: Generative AI guidance for lawyers
“ask what data the tool accesses and how it is trained, how will your input data be used and who owns input and output data”
- American Bar Association: Formal Opinion 512: Generative Artificial Intelligence Tools
“by their very nature, raise the risk that information relating to one client’s representation may be disclosed improperly”
Feedback and support submissions
Feedback and support can send a whole conversation through a separate route.
None of the sources checked speaks to this question directly.
Retention and zero-retention scope
Review each data category, feature and model. No-training is not no-retention.
What the sources say
- American Bar Association: Formal Opinion 512: Generative Artificial Intelligence Tools
“determine whether the [GAI tool] retains information submitted by the lawyer before and after the discontinuation of services”
- Solicitors Regulation Authority: Compliance tips for solicitors regarding the use of AI and technology
“in line with any timeframes specified in relevant data protection legislation”
Deletion and legal holds
Confirm deletion from live systems, indexes, backups and subprocessors, including exceptions.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“including data removal and deletion within the generative AI system”
- American Bar Association: Formal Opinion 512: Generative Artificial Intelligence Tools
“determine whether the [GAI tool] retains information submitted by the lawyer before and after the discontinuation of services”
Security, assurance and incidents
Access control, encryption, certification scope and incident handling.
The 4 questions in this groupShow the questions
Identity and matter access
Check permissions, tenant separation, least privilege, MFA, offboarding and connected sources.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“who has access to the data”
- American Bar Association: Formal Opinion 512: Generative Artificial Intelligence Tools
“investigate the [GAI tool’s] reliability, security measures, and policies”
Encryption and key control
Check transport, storage, keys and feature exceptions. Encryption does not answer lawful-access or output-quality questions.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“Understand how data is processed on the vendor side and the security measures they have in place to protect your data”
- New Zealand Law Society: Generative AI guidance for lawyers
“Ensure that robust security measures are in place to protect from data breaches”
Security assurance scope
A certificate must cover the entity, service, feature and period being assessed.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“regularly review the generative AI vendor’s data management, security and standards”
- New Zealand Law Society: Generative AI guidance for lawyers
“this includes being satisfied about what security measures the vendor has in place”
Incidents and continuity
Confirm notification, cooperation, records, escalation and usable fallback.
What the sources say
- American Bar Association: Formal Opinion 512: Generative Artificial Intelligence Tools
“that the lawyer will be notified in the event of a breach or service of process regarding production of client information”
- New Zealand Law Society: Generative AI guidance for lawyers
“Is there a plan in place in case the AI system fails?”
Contract, change and exit
The contract terms, changes to the service, and leaving.
The 3 questions in this groupShow the questions
Contract and processing terms
Check the actual contracting entity and accepted terms for the selected plan and model route.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“review the terms of service, especially regarding data rights, intellectual property, the geographic location of data and any liability clauses”
- Solicitors Regulation Authority: Compliance tips for solicitors regarding the use of AI and technology
“the liability between your firm and the supplier will typically be covered by the contract”
Model and supplier changes
Changes in models, features, terms, regions or recipients can require a fresh review.
None of the sources checked speaks to this question directly.
Exit and export
Preserve required matter records and evidence before account closure or migration.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“Consider whether you need or will have long-term support from a vendor, as well as an exit plan”
- New Zealand Law Society: Generative AI guidance for lawyers
“Consider how the firm can transition away from the tool, if needed.”
Agent actions and human review
What an agent may do on its own, and who checks the work.
The 2 questions in this groupShow the questions
Agent actions and approval
Distinguish reading information from sending, filing, editing or deleting it.
None of the sources checked speaks to this question directly.
Competent human review and fallback
A named reviewer needs relevant competence, access to original sources, time to check and a usable escalation or manual route.
What the sources say
- Law Society of England and Wales: Generative AI – the essentials
“review generative AI outputs for accuracy and factual correctness, including mitigation of biases and fact-checking”
- Law Society of England and Wales: Generative AI – the essentials
“implement a consistent process for checking and verifying the AI’s outputs for accuracy, reliability and liability”
The sources
- Law Society of England and Wales: Generative AI – the essentials (UK; checked 3 November 2026)
- Solicitors Regulation Authority: Compliance tips for solicitors regarding the use of AI and technology (UK, Updated 9 February 2026; checked 3 November 2026)
- American Bar Association: Formal Opinion 512: Generative Artificial Intelligence Tools (US, Issued 29 July 2024; checked 3 November 2026)
- New Zealand Law Society: Generative AI guidance for lawyers (NZ, Page updated 24 February 2026; checked 3 November 2026)
- Law Society of NSW, Legal Practice Board of WA and Victorian Legal Services Board + Commissioner (joint statement): Statement on the use of artificial intelligence in Australian legal practice (AU; checked 3 November 2026)
Using the questions
In the software workspace, a guided set of the same questions records what a vendor says, and a request PDF puts the questions to the vendor. Both show who has written about each question. Start from the software catalogue.
Nothing here is legal advice, and nothing approves a tool. The method page sets out what the framework does not decide.