QuestionAI GovernanceLegal ServicesVendor Risk

Can AI vendors use our client data to train their models?

3 July 2026
Answered by Rohit Parmar-Mistry

Short answer

Before putting client material into an AI vendor's system, check whether it may be used for training, evaluation or service improvement. Compare the contract and settings with client permissions and the firm's confidentiality and data-protection duties.

Detailed answer

Can AI vendors use client data to train their models?

Before putting client material into an AI vendor's system, check whether it may be used for training, evaluation or service improvement. Compare the contract and settings with client permissions and the firm's confidentiality and data-protection duties.

The source question asks whether customer data is used to train, fine-tune or improve AI models. Include it in a use-specific review before an assistant, search tool, knowledge base or workflow product receives professional data.

The safest answer is contractual, technical and evidenced

Before entering client data, establish the permitted use for that data, workflow and provider account. Compare contract terms, settings and available evidence with client permissions and confidentiality, data-protection and professional duties. Training may be prohibited for the use or require separate authorisation; an opt-out toggle is not the only possible control, and client permission alone does not resolve every duty. Do not rely on a sales answer as proof of the configuration.

If the vendor says data is not used for training, ask what counts as training, fine-tuning, model evaluation, abuse monitoring, human review, product analytics and service improvement. Check the relevant data processing terms, privacy notice, support policy and product documentation separately; a no-training statement does not necessarily cover the other uses.

Check AI vendor risk before client data enters the tool

What to ask the vendor before entering client data

Use plain questions that map to real controls:

  • Will any customer, client or end-user data be used to train, fine-tune or improve foundation models?
  • Is data used for evaluation, quality review, abuse monitoring or product analytics?
  • Can model-training use be disabled by default at tenant, workspace and user level?
  • Can the opt-out be written into the contract, not only selected in an admin screen?
  • How long are prompts, files, outputs, logs and embeddings retained?
  • What deletion process, backup schedule, derived-store limits and confirmation evidence are available, and what cannot be verified?
  • Which subprocessors can access the data and in which jurisdictions?
  • Will support staff or human reviewers see client content?
  • What audit logs can the firm export for regulator, client or insurer questions?

Why professional services firms need a higher bar

Professional-services uses may involve privileged material, regulated advice, special category data, transaction documents, claims files, audit evidence or commercial strategy. Use synthetic or expressly permitted offline cases for an initial test. Assess vendor evidence against the actual data and duties rather than assume every test contains client material or that sample text is harmless.

Have the responsible professional identify relevant duties: confidentiality, privilege and disclosure for legal work; any applicable Consumer Duty, SMCR, outsourcing and record-keeping requirements for financial-services work; and claims sensitivity, quality review and professional standards for insurance or accountancy. These are review topics, not a finding that each rule applies to every use.

How to turn the answer into an internal rule

Once the vendor response is clear, translate it into an internal rule that staff can actually follow. Do not leave people to interpret a 30-page data processing addendum each time they use a tool.

  • Approved for specified public or synthetic data: named demo or template uses only, after checking data rights, sensitivity and the test boundary.
  • Approved for specified client data with controls: named workflows and account configuration, with documented permitted purposes, training position, access, retention and proportionate evidence requirements. Training-disabled status alone is not clearance.
  • Not approved for client data: restrict use to separately approved general drafting or research data; do not enter confidential, personal, privileged or regulated material under that tier.
  • Blocked for the affected use: required training-purpose, retention, security or subprocessor evidence is missing or unsuitable. Record the gap and reviewer before reconsidering it.

Build an AI governance operating model that staff can use

Red flags in vendor answers

Be cautious when the vendor says data is not used for training but reserves broad rights for product improvement, evaluation or monitoring without explaining what data is involved. Also watch for opt-out controls that are available only on enterprise plans, undocumented admin settings, vague deletion commitments, missing subprocessor lists, or support policies that allow human review of content without a tight process.

A strong answer is specific. It identifies the data categories, purposes, retention, training position and any opt-out or other applicable restriction, subprocessors, security controls, deletion limits and available evidence. It also confirms whether the same rules apply to prompts, uploaded files, generated outputs, logs, embeddings and support tickets.

Conclusion

Do not enter client data until the specified use and account have been assessed and approved against the applicable permissions and duties. Record training and other processing purposes, restrictions, access, retention and available evidence. If required evidence is missing, stop the affected use; missing a technical opt-out does not by itself define whether another use or data category is safe.

Implement AI tools with the right data controls from the start

Need a safer AI route for your firm?

If this article reflects a live decision in your practice, the useful next step is to map the workflows, confidentiality risks, supplier controls and governance gaps before tools spread informally.

Frequently asked questions

Is a vendor privacy notice enough?

Not by itself. Compare the relevant contract, purposes, account settings and available evidence with the privacy notice and the applicable duties. Record inconsistencies and unresolved permissions before deciding on the use.

Can we rely on a model-training opt-out toggle?

Treat it as one control. Verify its scope, permissions, effective setting and current contract terms, including separate evaluation or monitoring uses. A toggle alone does not establish that client data is permitted.

Should staff paste client documents into public AI tools?

Only after the specific tool, account, workflow and data have been assessed and approved against applicable permissions and duties. Use synthetic or expressly permitted offline cases for testing; a public or private product label does not itself establish a safe boundary.

What if the vendor uses data for abuse monitoring?

Clarify what data is inspected, who can see it, how long it is retained and whether client content can be excluded or minimised.

Discuss your next step

Talk through your situation with Rohit, or take the assessment if you need a first read.