Salesforce expands Databricks partnership to add governance and access controls for AI agents

Salesforce is expanding its partnership with Databricks to make it easier for organizations to run AI agents on trusted enterprise data and, crucially, act on that data within existing security and approval rules. Announced at Databricks Data + AI Summit ’26, the effort focuses on closing a common enterprise gap: AI systems may be able to generate recommendations, but they often lack the permissions context and business-process controls required to execute actions safely.

The new plan builds on what Salesforce describes as an existing “Zero Copy” partnership between Salesforce Data 360 and Databricks Unity Catalog. Zero Copy is designed to let data be accessed across platforms without repeatedly ingesting and duplicating it into multiple systems. Salesforce and Databricks say the expanded collaboration adds governance capabilities intended to recreate security rules, identity handling, and access controls across both environments more consistently, reducing the risk and delay that can come from rebuilding permissions separately in each system.

Salesforce said the governance upgrades include Federated Authentication, planned identity mapping, governance interoperability, and metadata-aware access controls. The goal is to help users and AI agents understand what data they are authorized to access, which actions they are permitted to take, and how those actions align with enterprise sharing models and compliance requirements across Salesforce and Databricks.

Beyond identity and permissions, Salesforce and Databricks are also adding ways to discover relevant data and connect it to action. The companies announced new Federated Search capabilities intended to provide a single experience that is shared in both directions: Agentforce agents can search Databricks, while Databricks users can search Salesforce. Salesforce gave an example in which a sales representative’s agent could combine Salesforce account data with operational and usage information managed in Databricks to assess why a renewal might be at risk and then recommend next steps and initiate approved workflows.

Salesforce also highlighted additional integrations for agent ecosystem visibility and tool access. MuleSoft’s Agent Scanners for Databricks are described as providing automatic visibility in Agent Fabric, which is presented as the underlying layer where agent workflows and governance can be managed at scale. The companies also said they are exploring how Agentforce agents could be surfaced through Databricks Unity AI Gateway, linking discovery, reasoning, and controlled execution within the enterprise.

For where work happens, Salesforce said it is extending the partnership into Slack. New Slack integrations include Databricks Genie, the Databricks Q&A App, and Databricks Security App capabilities, which are intended to bring Databricks-powered insights, AI experiences, and security workflows directly into Slack channels. Salesforce cited scenarios such as a retail operations manager asking in Slack about stores at risk of inventory shortages due to weather-related shipping delays, with recommendations powered by Databricks Genie, or a healthcare security team getting alerted to anomalous access involving sensitive patient data through Databricks Lakewatch.

The announcement underscores how both companies are positioning AI agents as part of broader enterprise operations rather than standalone chatbots. Salesforce described the partnership as building a “shared foundation” so humans and AI agents can work side by side, combining enterprise business context with governance controls to reduce the chance of inappropriate access or unauthorized actions.

Executives tied the effort to customer feedback about trustworthy agent workflows. Databricks President Andy Kofoid said customers want AI agents to play a larger role in daily enterprise work, but that this requires access to trusted data, business context, and governance controls wherever the information lives. Salesforce President Rahul Auradkar echoed the same theme, saying the challenge is no longer building more agents, but giving them the trusted data, business context, governance, and workflows needed to operate safely at enterprise scale.

The companies did not provide pricing details or offer a universal timeline for every capability. Salesforce said availability is subject to customer agreements and can vary by region. It also listed a partial rollout schedule: MuleSoft Agent Scanner for Databricks and Data 360 foundational Zero Copy capabilities are generally available now, with Data 360 enhancements planned for H2 2026 and beyond. Salesforce added that the Slack Genie app is in public preview and is expected to reach general availability in H2 2026, while broader agentic search and open MCP-driven integrations are slated for H2 2026 and beyond.

For customers and the market, the practical watch items are whether these governance controls work as advertised across complex identity and data-sharing setups, and how quickly enterprises can operationalize agent workflows that cross both Salesforce applications and Databricks data platforms. The partnership also indicates where agent platforms are heading: toward permission-aware execution and tighter integration into everyday work tools like Slack, rather than relying on a single system’s internal access model. Next, Salesforce and Databricks are likely to focus on rollout coverage, integration depth, and measurable customer use cases that demonstrate safer, faster time to action.