playbook
ChatGPT connectors and customer data: small-team security checklist
How small teams should approve ChatGPT connectors to Gmail, Drive, Slack, GitHub, CRM, and other systems before customer data becomes searchable.
Bottom line
ChatGPT connectors are higher risk than copy-paste prompts. A prompt exposes what an employee enters. A connector can expose email, files, tickets, code, calendars, or customer records that the employee can access in another system.
Small teams should approve connectors as separate workflows, even when ChatGPT Business or Enterprise is already approved.
Pair this page with the broader customer data in ChatGPT playbook.
Connector approval matrix
| Connector target | Typical data exposed | Default posture |
|---|---|---|
| Google Drive or Microsoft 365 | Docs, spreadsheets, decks, customer files, board materials | Pilot only with named users and folder rules. |
| Gmail or Outlook | Customer messages, attachments, contracts, invoices, personal data | Restrict until use case and retention are documented. |
| Slack or Teams | Internal discussions, customer escalations, incident threads | Restrict by workspace/channel rules. |
| GitHub | Source code, issues, PRs, security discussions | Engineering approval required. |
| CRM or helpdesk | Customer records, tickets, account history, payment context | High-risk workflow approval required. |
| Calendar | Meeting metadata, invitees, customer names, confidential events | Allow only when the use case needs it. |
The right question is not “does the connector work?” The right question is “what can this user already access, and is it appropriate for ChatGPT to retrieve it for this use case?”
Minimum approval record
Use this template before enabling a connector:
ChatGPT connector approval
Connector:
Workspace:
Business owner:
Admin owner:
Users or group:
Business purpose:
Data sources:
Customer data involved:
Restricted folders, channels, labels, or repos:
Allowed prompt types:
Prohibited prompt types:
Retention expectations:
Logging and review owner:
Approval date:
Renewal date:If you cannot name the owner and the data source, do not enable the connector.
Admin checks
OpenAI’s connector documentation says each user authorizes their own account and ChatGPT accesses content within that user’s existing permissions. That is useful, but it also means source-system permissions become the security boundary.
Before rollout:
- Review source-system groups and permissions.
- Remove stale users before enabling connectors.
- Separate customer folders, internal planning folders, and restricted executive/legal folders.
- Disable broad sharing links where possible.
- Decide whether synced data is acceptable for the workspace and region.
- Confirm app calls and compliance logs are available to the right admins where plan support exists.
- Document the offboarding step for each connected system.
Do not approve a connector to fix messy permissions. Fix permissions first.
Customer-data prompt rules
For connector-backed workflows, employees should not ask broad questions such as:
Summarize everything about Customer A from Drive, Gmail, and Slack.Use scoped prompts:
Using only the approved support folder and the last 30 days of tickets for Customer A, draft a support escalation summary. Do not include personal billing details, contract terms, or unrelated customer names.Even better, use a pre-approved workflow where the source, data class, and output destination are known.
Rollout sequence
| Phase | Action |
|---|---|
| Inventory | List every requested connector and source system. |
| Permission cleanup | Remove stale users, broad groups, and old shared folders. |
| Pilot | Enable one connector for one low-risk use case and a small user group. |
| Review | Check logs, sample outputs, and employee prompts. |
| Expand | Add only the next specific use case, not every connector at once. |
| Renew | Reapprove connectors monthly while the program is new. |
Evidence checked
- OpenAI business data privacy, security, and compliance
- OpenAI admin controls, security, and compliance in apps and connectors
- OpenAI Data Controls FAQ
- Can employees paste customer data into ChatGPT?
- Small Team AI Security Checklist
FAQ
Are connectors safer than pasted customer data?
Not automatically. Connectors can reduce manual copying, but they can also retrieve more records than the employee intended. Treat connector approval as a higher-risk decision.
Does ChatGPT train on connector data in Business workspaces?
OpenAI says information accessed from apps is not used to train models for ChatGPT Business, Enterprise, and Edu customers. That does not remove retention, access, logging, contract, or customer-trust concerns.
What is the first connector a small team should approve?
Pick the lowest-risk connector tied to a clear workflow. For many teams, that is a restricted documentation folder, not email, Slack, CRM, or helpdesk.
Recommended next step
Run the AI Tool Risk Checker for the proposed connector workflow before enabling it for a whole team.