OpenAI announced a new "Company Knowledge" feature for ChatGPT Enterprise and Edu on 23 October 2025, allowing organisations to connect internal knowledge bases directly to the model [source]. The feature indexes company documents, wikis, and shared drives, making them searchable through natural language queries in ChatGPT.

The announcement describes automatic syncing with platforms including Google Drive, Microsoft SharePoint, and Confluence. Administrators control which repositories are indexed and which users have access. OpenAI states that company data is not used to train its models and that access controls mirror existing permissions in source systems.

The feature introduces new failure modes for enterprise deployments. If the indexing system misinterprets access permissions, users could receive answers derived from documents they should not see. If the model hallucinates citations or conflates similar document names, employees may act on incorrect information attributed to authoritative internal sources. Synchronisation lag between source systems and the ChatGPT index could surface outdated policies or procedures.

OpenAI provides no public documentation of how conflicts between company knowledge and the model's pre-training are resolved when answers differ. The announcement does not specify retention periods for indexed data or describe how deletions in source systems propagate to ChatGPT's index.

Enterprise customers using this feature should verify that cited documents match actual source content and that access controls function as intended. The feature is rolling out to ChatGPT Enterprise and Edu customers beginning 23 October 2025, with broader availability planned for early 2026.