What actually leaves the firm when AI is used
Trace one AI workflow from the user input through uploads, output and logging. Check where each category is processed, retained and accessible before approving client data.
Record what moves, where processing occurs, who has access and how long each category is retained. Compare supplier statements with settings and records so the firm can explain the flow to a client, insurer or internal reviewer.
The safe answer: map prompts, files, outputs, logs and support access
Check five parts of the proposed flow. First, prompts and instructions: text staff type, paste or dictate. Second, uploaded material: PDFs, spreadsheets, emails, contracts, case notes, call transcripts and screenshots. Third, outputs: drafts, summaries, recommendations, classifications and generated analysis. Fourth, metadata and logs: check whether user IDs, timestamps, IP addresses, usage analytics, moderation events, error traces or audit logs are collected and where they go. Fifth, human or subprocessor access: identify any vendor support, model-provider, cloud-processing or security-review access that the service allows. Mark each part as present, absent or unresolved; do not assume every item leaves the firm.
For confidential, privileged, personal or commercially sensitive information, assess the task's permitted use and required controls before approval. Check minimisation, retention, access, output review and exceptions against that flow rather than applying one control set to every tool.
Map AI data exposure before rollout
What to ask the vendor before approving the tool
Ask for plain evidence rather than broad assurances. The vendor should explain whether prompts, uploaded files and outputs are stored, used for training, reviewed by humans, shared with model providers, included in telemetry, or retained in backups. For regulated or client-facing work, the answer should be specific enough for procurement, compliance and delivery owners to act on.
Useful checks include: where processing occurs, whether data is encrypted in transit and at rest, whether customer data is logically separated, whether retention can be shortened or disabled, whether logs can be exported, whether support access is controlled, and whether subprocessors are listed. If the tool is embedded inside another platform, check both the application vendor and the underlying model or cloud provider.
How firms should set internal rules
Policies should turn the data map into day-to-day rules. Staff need to know which tools are approved, which data can be entered, which work must stay out of public or consumer AI tools, and when a human reviewer must sign off the result. A good rule is to treat client-identifiable data, special category personal data, privileged material, unpublished financial information and confidential strategy as restricted unless a named owner has approved the exact workflow.
For client work, decide which records are needed to explain the tool, data category, reviewer, decision and redactions or other controls. Check those records against the work performed; their existence does not establish that the output was correct or the use permitted.
Keep AI governance evidence current
A practical approval checklist
Prepare an approval record for the use case, data categories, supplier controls, retention, access, required review and escalation. For higher-risk work, assess a test with synthetic or appropriately redacted material and inspect the resulting logs. If the test leaves a material flow unexplained, keep that use unapproved while the owner investigates.
Set the review depth from the information, task and effect of the output. Internal drafting with non-sensitive inputs may need different checks from work that informs client advice or financial decisions; the task label alone does not establish its risk.
Conclusion
Map prompts, files and outputs alongside any metadata, logs, subprocessor processing and support access present in the workflow. Use the observed flow and unresolved questions to decide the permitted data, review and records for that use. A completed map is not approval by itself.
Build controlled AI workflows for client work
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.
