Learn · Industry Guides

AI and client confidentiality

Short answer. Professional confidentiality duties apply whatever tool is used. Firms can use AI with client information if they choose tools with suitable terms, keep clients strictly separate, limit what AI receives, keep a person responsible for output and keep a record of access.

Last reviewed 24 September 2026

The obligation has not changed

Accountants, solicitors, financial advisers and consultants hold information in confidence for their clients. Professional bodies such as the ICAEW and the SRA set expectations around confidentiality, and client contracts and engagement letters often add their own terms. UK GDPR applies to the personal data inside client files.

None of this changes because a tool is new. The question is how to meet the same obligations when AI is involved. This page is general information, not legal or regulatory advice; check your own professional body’s current guidance.

Where AI creates confidentiality risk

  • Disclosure to a provider. Information sent to an AI service is processed by that provider. Whether that is acceptable depends on the provider’s terms and your obligations to the client.
  • Mixing clients. An AI with access to many clients’ files can blend them, or retrieve the wrong one.
  • Over-broad retrieval. A search across a whole document store may surface files the person working on the task should not see.
  • Leakage in output. A draft for one client that mentions another.
  • Unapproved tools. Staff using personal accounts outside the firm’s control.
  • External sharing. Clients or third parties receiving more than intended.

Practical safeguards

Choose tools deliberately

Use AI services on business terms you have reviewed: retention, training use, processing location and subprocessors. Check your provider’s current settings, as terms vary by plan.

Separate clients by design

Work on one client should draw only on that client’s information. This is easier to enforce in the system than to remember in practice.

Minimise what AI receives

Provide the portion of a client’s record the task needs. Internal risk assessments, AML notes, fee discussions and staff comments rarely need to be included.

Keep a responsible person

AI-prepared work should be reviewed by someone who takes professional responsibility for it. Require approval before AI-drafted communication reaches a client.

Keep records

Record which client information AI accessed, for whom and what it produced.

Tell clients

Consider whether your engagement letters or privacy notices should describe your use of AI. Many firms find that clients prefer clarity.

Questions for your firm

  • Which AI tools are approved for client work, and on what terms?
  • Can a member of staff’s AI reach clients they do not work on?
  • Which categories of client information are excluded from AI?
  • Who approves AI-drafted client communication?
  • Could you show a client what AI accessed about them?

Clients using their own AI

Clients increasingly want to use their own AI with information from their adviser. That is manageable if you share a defined portion of their record, not your workspace, and if you can see and end that access.

How Kroy approaches it

Kroy keeps client boundaries at the point of access. Each request is checked against the person, the agent, the client entity, and your policies, so an employee’s AI receives only the client information they are entitled to. Internal notes can be excluded from every AI projection. Client-facing sharing uses guest access with its own projection, expiry and audit trail.