An AI request can include more than text: attachments, conversation history, connected accounts, or information copied from another system. Before sending it, users should understand what data is involved, which service receives it, and whether its use follows their organization’s policies.
This article offers general awareness and practical habits. It is not legal advice or the privacy policy of any specific service. Data handling varies by product, plan, and configuration; consult the relevant official documentation.
Recognize sensitive data
Sensitive data is not limited to passwords. Depending on the context, identifying information, customer records, financial data, health information, unpublished contracts, trade secrets, or internal material may need protection.
A passage can remain sensitive even after names are removed. A combination of role, location, dates, and events may still identify a person or organization. Consider the full context, not just an individual field.
Check before sending
Before pasting content or uploading a file, ask:
- Is real data necessary? Could a sample or anonymized version work?
- Which service will receive it? Is it approved by your organization, or a personal account?
- How is the data handled? Check retention settings and the policy for the specific plan.
- Who can access it? Who can open the file, and are their permissions broader than needed?
- Can access be removed? Know how to handle the data when the task is done.
If you do not know the answer to an important question, pause and ask the data owner or an authorized person in your organization.
Attachments need review too
Before uploading a file, check for hidden pages, comments, revision history, contact details, metadata, or other people’s data. A spreadsheet may contain hidden sheets; a document may retain comments or an author name.
Provide only the portion needed. If AI needs to analyze a few columns, you may not need to send an entire database. Reducing the input can limit exposure and keep the task focused.
Connected accounts and permissions
A tool connected to email, cloud storage, or a task system may access data within the permissions granted. “Connected” does not mean every action is appropriate for every task.
Prefer the minimum access needed. Separate read and write access where possible; review old connections and revoke those no longer used. For sending, deleting, publishing, or changing data, require human confirmation first.
API keys and credentials
API keys, passwords, recovery codes, and access tokens are credentials. Do not put them in prompts, public documents, screenshots, or shared code. If AI needs to inspect code, replace secrets with dummy values and check again before sharing.
If you suspect a key was exposed, follow the relevant service’s process to revoke or rotate it and notify the responsible person. Deleting a message in an interface does not necessarily remove the information everywhere.
Local-first is not an absolute privacy guarantee
An application may process some data on a device and still connect to another service for certain tasks. “Local-first” by itself does not prove that all data stays on the computer, never travels over a network, or is never stored elsewhere.
Verify the actual data flow: which operations happen locally, which call external services, which providers receive content, and which settings affect retention. Do not infer a privacy commitment from a marketing label.
Build safe habits as a team
Organizations can agree on approved tools, permitted data types, anonymization practices, and reviewers. Guidance should include clear examples, a reporting channel, and a date for review.
Users should also check outputs before sharing, because AI may repeat input data or place it somewhere unexpected. Privacy does not end with upload; it also involves where drafts are stored and who receives the result.
Check again after the work is done
Finishing a task does not necessarily end the data lifecycle. Check where drafts are stored, who can open them, whether sharing links remain active, and whether the material still needs to be kept. A summary based on sensitive information may reveal important details even after names are removed.
In a team, agree where approved results belong and how to distinguish them from drafts. Avoid creating copies across multiple services just for convenience. Follow the service and organization’s process when deleting data or revoking access; do not assume removing one visible copy removes all copies.
Key takeaway: Share only necessary data, with an appropriate service and limited permissions. If you do not understand how the information is handled, use sample data or ask the owner first.
In summary
AI can help with documents and data, but users must consider the information type, accounts, connected permissions, and data flow. No single setting applies to every provider.
For sensitive tasks, follow your organization’s policy and the official documentation for the service. If real data has not been approved for use, do not submit it.