Data Sovereignty is the ability of an organization to keep real control over the data it owns, typically stored on its own servers, meaning it can keep that data available under its own rules and decide which software may access it. That excludes organizations storing their data in third-party systems like SaaS (Software as a Service), which grant access under the vendor's rules rather than the organization's, an arrangement too common among companies that rely on rented software to hold their records.
For example, a company whose customer list sits in a rented sales application can read it only in the formats the vendor allows. Some consider that keeping data in a vendor's system equals renting it rather than owning it, because the organization is exposed to the restrictions of a third party and dependent on that party's commercial interests. This is becoming increasingly relevant with AI agents, because reading company data, or the system of record, is what an agentic system does before it can do anything else.
Salesforce revised the developer terms for its workplace messaging product Slack in May 2025, prohibiting bulk export and barring lasting copies or indexes of Slack data, which cut off Glean, an enterprise search tool that had indexed Slack messages alongside a hundred other applications. SAP, the German company whose software runs the accounts and supply chains of many large manufacturers, changed its interface policy in April 2026 to bar outside AI programs from running their own sequences of requests against customer data. ServiceNow now routes outside agents through a layer called Action Fabric, priced by the number of operations each one completes. Much of the industry reads these changes as a way of pushing customers toward the vendor's own AI products, rather than as a way of serving customers with data those customers already own.
Some suppliers reject that description. Christian Klein, chief executive of SAP, told investors in April 2026 that customers would not be charged for reaching their own data and that the company intends to keep its architecture open to programs it did not build.
.png)
In work automation, direct access to company data is a core requirement, because an AI agent has to read a record, act on it, and read it again as the task proceeds, so the terms attached to Retrieval set what can be automated and at what price. Microsoft began charging for that access on 16 June 2026, billing an outside agent that reads Microsoft 365 mail, files, or calendars at 0.1 Copilot Credits for every tool call. Datadog capped outside agents at 5,000 requests a day instead. For example, an agent that rereads the same dashboard on every loop reaches that limit before the afternoon, which forces the person who built it to store results between runs.
