HubSpot has moved a sensitive-data permission for AI connectors into public beta, according to its October 2026 changelog.
HubSpot previously blocked AI connectors from reading Sensitive Data properties and engagements in accounts with Sensitive Data turned on. In the new public beta, ChatGPT, Claude, and Copilot can be granted access to Sensitive Data properties and engagements. Gemini and HubSpot’s Remote MCP server can be granted access to engagements only.
HubSpot’s setup document says the beta applies to accounts with Sensitive Data settings turned on and lists Enterprise-level HubSpot subscriptions, including Marketing Hub Enterprise, Sales Hub Enterprise, Service Hub Enterprise, Data Hub Enterprise, Content Hub Enterprise, Smart CRM Enterprise, and Revenue Hub Enterprise.
Super Admins must opt the account into the beta, then configure optional permissions during installation or reauthentication of a supported connector. If a connector is already connected, HubSpot says it needs to be disconnected from both the AI provider and HubSpot, then reauthenticated before the new permissions can be configured.
The field-level change is only part of the issue
HubSpot’s permissions table also matters because it covers engagements: calls, emails, meetings, notes, and tasks. Those records can carry far more context than a structured CRM field, especially in sales, support, healthcare-adjacent, financial-services, or professional-services accounts.
That broader context can make AI analysis more useful. It also means permission reviews need to cover conversation history and activity records, not only custom fields marked as sensitive.
What remains blocked
HubSpot says AI connector access is limited to standard Sensitive Data properties, including properties that contain HIPAA-protected health data, and engagement data. Highly Sensitive Data properties remain restricted.
HubSpot’s Sensitive Data documentation describes Sensitive Data as information such as personal identification, financial, health, or medical information. Highly Sensitive Data includes examples such as social security numbers and receives tighter restrictions in HubSpot tools.
HubSpot also says access depends on app permissions and the user’s existing CRM object or property-level permissions, which the beta does not override.
Connector documentation is still catching up
The dedicated beta page says supported AI connectors can be granted access to Sensitive Data, but several connector-specific Knowledge Base pages still contained older no-access language as of November 4, 2026.
HubSpot’s ChatGPT and Claude connector pages, for example, still included notes saying those connectors could not access custom Sensitive Data properties or engagement data when Sensitive Data was turned on. Copilot and Gemini pages also described no Sensitive Data property access in their connector-specific FAQs.
That mismatch does not necessarily mean the beta is unavailable. It does mean administrators should rely on the beta access page, the in-product permission screen, and their own test account behavior before assuming a connector has or does not have access.
Third-party AI terms still matter
Once HubSpot data is sent through a third-party AI connector, the AI provider’s own terms, retention practices, and workspace controls matter. HubSpot’s ChatGPT connector documentation says OpenAI retains data accessed through the connector to make the connector work in ChatGPT, and that data retention and deletion can also be governed by the third-party tool’s terms.
OpenAI’s connected-apps documentation says app permissions in ChatGPT do not grant an app new access by themselves. The available data and actions are determined by the app, the access granted when connected, and workspace controls. That distinction matters here because HubSpot is adding optional connector permissions at the CRM side of the connection.
For managed ChatGPT workspaces, OpenAI’s data controls documentation says Business, Enterprise, Edu, and Healthcare content is not used for model training by default. Personal account settings and third-party plan terms can differ, which makes account type and workspace governance part of the risk review.
What administrators should review before enabling the beta
For administrators, this beta should be treated as a data-governance change, not only an AI productivity feature. The access is read-only, but read-only access can still expose sensitive context to another system and to any authorized user who can query it.
Before enabling the beta, organizations should identify which sensitive fields and engagement types are needed for AI workflows, confirm who has Super Admin or App Marketplace permissions, review which users can install each connector, and check the third-party AI plan terms that apply. Audit logs and approval workflows should also be reviewed before access expands beyond a small test group.
HubSpot says new AI connector permissions may take as long as two hours to appear after reconnection. That gives administrators a natural point to test with a narrow data set before moving the beta into broader use.
HubSpot’s move reflects a broader shift in business software: AI tools are becoming interfaces for CRM data, not just writing assistants or research tools.
I can suggest related Tech Help Canada articles based on the topic you’re reading now.
Tech Help Canada Staff researches, writes, and reviews practical content for business owners and professionals. Our coverage spans business, marketing, SEO, technology, and the tools and systems people use to grow and operate online. We focus on clear, useful information backed by research, hands-on experience, and editorial review. Learn more about our team and editorial standards. Need help with something? Contact Us
