Lakeflow Connect added managed connectors for OpenAI and Anthropic, landing AI provider data in Unity Catalog

The most concrete OpenAI-related release in October 2026 came on the 26th, when Lakeflow Connect added native, fully managed connectors for AI providers including OpenAI and Anthropic. The same release covered Atlassian, Microsoft SharePoint, Google Workspace and GitHub. Data from these sources, structured and unstructured, lands directly in Unity Catalog, so it falls under the same governance as other tables and needs no infrastructure to run [1].

Governance of external models was the other main thread. On October 1, Databricks announced general availability of the Unity AI gateway, which provides centralized governance, cost controls and observability for agents that call multiple models [7]. A community post from October 19 summarized a large set of Unity Gateway updates shipped in September [3]. A practitioner report on October 7 described a limit: managed Omnigent could not use external model providers such as Azure OpenAI or Gemini through Unity Gateway [5]. The post did not describe a resolution, so readers should check current documentation for the present state.

On the tooling side, MLflow 3.17.0 was released on October 7. It added fine-grained access control for runs and traces, linked evaluation metrics to managed datasets, and supported Databricks Unity Catalog model service resources. Teams running self-hosted SQL tracking servers had to run a coordinated schema migration with mlflow db upgrade and stop all writers first, because rolling deployments were not supported [6].

Industry coverage touched OpenAI as well. A September roundup video published on October 9 discussed OpenAI DevDay announcements, including persistent agents called dots, alongside the general availability of the Genie one Model Context Protocol [4]. A community article on October 23 looked at designing persistent memory for production agents [2].