Why client data in AI tools needs a clear answer
What happens to client data depends on the exact tool, account settings and agreement. Before use, check processing, access, retention, onward use and which client materials the approved workflow permits.
Check the tool, account, contract and the firm's operating rules together. Do not infer a data processing agreement, retention controls or audit logs from an enterprise label; establish which terms and features apply to the account staff will use.
The safest practical answer for client data
Client data should only go into AI tools that the firm has approved for that type of information, after checking data processing terms, retention settings, training use, access controls, location of processing, auditability and human quality review. If those points are unknown, the safe rule is: do not enter confidential, privileged, personal or commercially sensitive client material.
Give the engagement's relevant reviewers the tool, task, material and proposed recipients. Record their applicability decision, supporting evidence and remaining restrictions so staff can distinguish approved use from mere technical access.
Map where client data touches AI tools
What can happen to the data
Trace whether the workflow transmits client material to supplier systems and how it is processed. Check the applicable retention, safety or abuse monitoring, human access and model-training terms and settings. Record each as established, absent or unresolved rather than assuming a typical service behaviour.
Compare training exclusion, retention, access, processing regions, subprocessors and logging against the exact service and account. Request evidence for the commitments the firm needs; neither a brand name nor a legal-specific or enterprise description establishes that those commitments apply.
The governance checks before staff use AI with client material
Before approving an AI tool for client work, the firm should answer these questions in writing:
- What categories of client data may be entered, and which are prohibited?
- Is the provider acting as a processor, controller or independent service under the relevant terms?
- Is there a data processing agreement, and does it cover subprocessors, security, retention and deletion?
- Can prompts or uploaded files be used for training, safety review or product improvement?
- In which regions is the data processed and stored, and can the firm evidence that position?
- What logs, admin controls and review trails are available if a client or regulator asks?
- Who signs off outputs before they influence client advice, filings, valuations, reports or communications?
Turn the recorded decisions into staff instructions: approved tools and accounts, permitted data, required review and the exception route. Check those instructions against the workflow rather than treating the written policy as proof of control.
Put AI usage rules under active governance
How this applies in legal and regulated professional services
For the engagement, ask the responsible reviewers to establish the applicable confidentiality, privilege, professional and personal-data requirements. Include purposes, recipients, locations and security in the evidence they receive. Record their conclusions rather than presenting the checklist as SRA or UK GDPR compliance.
Assess the task and information, not just the tool. Compare a generic checklist using public information with a client's evidence bundle, contract, claim chronology, financial model or board paper, and record the permissions and review needed for each proposed use.
A workable policy for teams
A practical policy can be short, but it must be specific. It should name approved tools, permitted data classes, prohibited material, required redaction, review steps, logging expectations and escalation routes. It should also explain what staff should do when a client asks whether AI was used.
Inventory the tools in use, users, data, purposes and applicable terms. Use the record to identify unresolved uses and controls to check before expanding access; the inventory itself does not establish safe use.
Conclusion
Decide client-data permissions from the tool and account, data classification, contract review, relevant records and output-review responsibilities. Where required evidence or approval is missing, pause the affected use and assign the unresolved question to its owner.
Implement a controlled AI workflow
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.
