Prompt privacy: treat generative AI as a new data-egress channel

Generative AI adds a new data-egress path to the workplace: the prompt box.

A user can paste a customer complaint, a contract, a spreadsheet or a medical note into an external service in seconds. That action may become a personal-data processing operation.

Start with data minimization

Ask whether the task needs real identities at all.

For rewriting, summarizing or translating, the model often needs the facts, not the person's name, phone number or account ID.

Redaction is not always anonymization

Removing a name does not automatically make a document anonymous. Context, job title, dates, location or rare circumstances may still identify a person.

Check the account and service terms

A managed enterprise plan may provide different retention, training and administrative controls than a personal account.

Verify:

• prompt retention;
• model-training settings;
• subprocessors;
• data location;
• deletion options;
• admin controls;
• contractual terms.

Block obvious high-risk inputs

Create clear rules for health records, payroll, disciplinary files, IDs, credentials, API keys and confidential documents.

Files deserve extra caution

Documents may contain hidden sheets, comments, metadata, signatures or third-party data that the user did not intend to share.

Govern the output too

AI can infer, classify or invent information about people. Human review remains essential before generated content is placed into a customer file, HR record or business decision.

A practical AI policy is not anti-AI. It makes adoption safer by defining approved tools, permitted data and escalation rules.

Full Spanish guide: RGPD, prompts and generative AI in companies.