Workplace security awareness and email review
AI security awareness

Protecting data when using AI

AI can move information beyond the document or conversation where it began. Understand the service, settings and connections before sharing sensitive material. Compass is available now; this is guidance rather than legal advice, and describes no data-protection feature.

Check the information before sharing it

Review prompts, files and screenshots for personal data, confidential plans, client details and credentials. Consider hidden spreadsheet tabs, comments and metadata. Follow classification and handling rules rather than judging only whether the task seems routine. Conversation history may also contain information entered for an earlier task.

Share the minimum necessary for the approved purpose. Prefer synthetic examples or reduced datasets where suitable. Removing names does not necessarily anonymise information; dates, locations and unusual events can identify someone. Review the complete context rather than treating a simple name-removal exercise as sufficient.

Understand retention and contractual boundaries

Check the exact service and account terms. Establish retention, training use, service-personnel access and deletion exceptions. Closing a conversation may not remove every retained copy. Disabling training and disabling storage are different settings and may have different availability across subscriptions.

Assess UK GDPR requirements with appropriate expertise where personal data is involved, including lawful basis, transparency, processor arrangements, transfers and impact assessment where required. Supplier security claims do not establish that your use is lawful or appropriate. Record uncertainty and resolve it before exposing sensitive records.

Review connectors and downstream access

A connection to email, storage or business applications may retrieve more than a user manually enters. Review accessible locations, records and identities. Apply least privilege rather than organisation-wide access for convenience. The connector's identity and permissions need the same scrutiny as the visible AI account.

Check where outputs go and who can see them. Summaries can reproduce source data; connected actions can send it elsewhere. Require authorisation for sharing and review after role or integration changes. A restricted source folder is not enough if the generated response is published to an unrestricted channel.

Prepare a response to accidental disclosure

Report an upload to an unapproved service promptly. Record the service, account, time and type of information. Avoid reproducing the exposed material in extra tickets unless responders need it. If credentials were included, tell the technical team what kind without pasting them again.

Authorised responders assess containment, provider contact, credential changes and notification obligations. Deleting a visible chat may not resolve retained copies. Support the reporter and address process weaknesses. Compass is available now; awareness and reporting complement enforceable controls rather than replace them.

Questions

Frequently asked questions

Does disabling training mean nothing is retained?

Not necessarily. Training use and retention are separate questions. Check applicable settings, terms and exceptions.

Is anonymisation just removing names?

No. Other details can identify people. Assess re-identification risk and seek specialist advice where appropriate.

Should I delete the chat and say nothing?

No. Report promptly so responders can assess the information and retention. Deletion alone may not contain the exposure.

Discuss your staff awareness priorities.

Tell us which situations staff are struggling with and what you already have in place. Compass is available now; we can discuss how it would fit your requirements.

Discuss awareness requirements